Loading content...
Loading GPU specifications...
Loading GPU benchmarks...
Quick Answer: NVIDIA H200 SXM 141GB offers 141GB VRAM and starts around current market pricing. It delivers approximately 909 tokens/sec on unsloth/Llama-3.2-1B-Instruct. It typically draws 700W under load.
This GPU offers reliable throughput for local AI workloads. Pair it with the right model quantization to hit your desired tokens/sec, and monitor prices below to catch the best deal.
Buy directly on Amazon with fast shipping and reliable customer service.
| Model | Quantization | Tokens/sec | VRAM used |
|---|---|---|---|
| unsloth/Llama-3.2-1B-Instruct | Q4 | 909.11 tok/sEstimated Auto-generated benchmark | 1GB |
| Qwen/Qwen2.5-3B-Instruct | Q4 | 907.79 tok/sEstimated Auto-generated benchmark | 2GB |
| google/gemma-2b | Q4 | 895.08 tok/sEstimated Auto-generated benchmark | 1GB |
| unsloth/Llama-3.2-3B-Instruct | Q4 | 892.67 tok/sEstimated Auto-generated benchmark | 2GB |
| ibm-granite/granite-3.3-2b-instruct | Q4 | 891.91 tok/sEstimated Auto-generated benchmark | 1GB |
| google-bert/bert-base-uncased | Q4 | 885.47 tok/sEstimated Auto-generated benchmark | 1GB |
| nari-labs/Dia2-2B | Q4 | 875.90 tok/sEstimated Auto-generated benchmark | 2GB |
| allenai/OLMo-2-0425-1B | Q4 | 866.34 tok/sEstimated Auto-generated benchmark | 1GB |
| google/embeddinggemma-300m | Q4 | 856.73 tok/sEstimated Auto-generated benchmark | 1GB |
| WeiboAI/VibeThinker-1.5B | Q4 | 855.99 tok/sEstimated Auto-generated benchmark | 1GB |
| google-t5/t5-3b | Q4 | 851.74 tok/sEstimated Auto-generated benchmark | 2GB |
| context-labs/meta-llama-Llama-3.2-3B-Instruct-FP16 | Q4 | 850.99 tok/sEstimated Auto-generated benchmark | 2GB |
| google/gemma-2-2b-it | Q4 | 848.94 tok/sEstimated Auto-generated benchmark | 1GB |
| google/gemma-3-1b-it | Q4 | 847.64 tok/sEstimated Auto-generated benchmark | 1GB |
| inference-net/Schematron-3B | Q4 | 834.31 tok/sEstimated Auto-generated benchmark | 2GB |
| apple/OpenELM-1_1B-Instruct | Q4 | 833.64 tok/sEstimated Auto-generated benchmark | 1GB |
| meta-llama/Llama-Guard-3-1B | Q4 | 831.69 tok/sEstimated Auto-generated benchmark | 1GB |
| meta-llama/Llama-3.2-3B | Q4 | 831.65 tok/sEstimated Auto-generated benchmark | 2GB |
| meta-llama/Llama-3.2-1B-Instruct | Q4 | 830.93 tok/sEstimated Auto-generated benchmark | 1GB |
| tencent/HunyuanOCR | Q4 | 828.14 tok/sEstimated Auto-generated benchmark | 1GB |
| meta-llama/Llama-3.2-1B | Q4 | 825.66 tok/sEstimated Auto-generated benchmark | 1GB |
| bigcode/starcoder2-3b | Q4 | 809.33 tok/sEstimated Auto-generated benchmark | 2GB |
| unsloth/gemma-3-1b-it | Q4 | 805.20 tok/sEstimated Auto-generated benchmark | 1GB |
| deepseek-ai/DeepSeek-OCR | Q4 | 800.62 tok/sEstimated Auto-generated benchmark | 2GB |
| ibm-research/PowerMoE-3b | Q4 | 796.60 tok/sEstimated Auto-generated benchmark | 2GB |
| LiquidAI/LFM2-1.2B | Q4 | 786.08 tok/sEstimated Auto-generated benchmark | 1GB |
| facebook/sam3 | Q4 | 781.17 tok/sEstimated Auto-generated benchmark | 1GB |
| TinyLlama/TinyLlama-1.1B-Chat-v1.0 | Q4 | 776.34 tok/sEstimated Auto-generated benchmark | 1GB |
| meta-llama/Llama-3.2-3B-Instruct | Q4 | 774.84 tok/sEstimated Auto-generated benchmark | 2GB |
| Qwen/Qwen2.5-3B | Q4 | 764.55 tok/sEstimated Auto-generated benchmark | 2GB |
| deepseek-ai/deepseek-coder-1.3b-instruct | Q4 | 759.45 tok/sEstimated Auto-generated benchmark | 2GB |
| Qwen/Qwen-Image-Edit-2509 | Q4 | 757.77 tok/sEstimated Auto-generated benchmark | 4GB |
| huggyllama/llama-7b | Q4 | 756.57 tok/sEstimated Auto-generated benchmark | 4GB |
| Qwen/Qwen2.5-Coder-7B-Instruct | Q4 | 756.47 tok/sEstimated Auto-generated benchmark | 4GB |
| swiss-ai/Apertus-8B-Instruct-2509 | Q4 | 756.26 tok/sEstimated Auto-generated benchmark | 4GB |
| lmstudio-community/Qwen3-4B-Thinking-2507-MLX-6bit | Q4 | 756.06 tok/sEstimated Auto-generated benchmark | 2GB |
| trl-internal-testing/tiny-Qwen2ForCausalLM-2.5 | Q4 | 755.39 tok/sEstimated Auto-generated benchmark | 4GB |
| petals-team/StableBeluga2 | Q4 | 753.22 tok/sEstimated Auto-generated benchmark | 4GB |
| meta-llama/Llama-2-7b-hf | Q4 | 751.92 tok/sEstimated Auto-generated benchmark | 4GB |
| deepseek-ai/DeepSeek-R1-Distill-Llama-8B | Q4 | 749.64 tok/sEstimated Auto-generated benchmark | 4GB |
| Qwen/Qwen2.5-Coder-1.5B | Q4 | 749.08 tok/sEstimated Auto-generated benchmark | 3GB |
| microsoft/Phi-4-multimodal-instruct | Q4 | 748.81 tok/sEstimated Auto-generated benchmark | 4GB |
| Qwen/Qwen3-4B-Instruct-2507 | Q4 | 745.01 tok/sEstimated Auto-generated benchmark | 2GB |
| Qwen/Qwen3-0.6B-Base | Q4 | 744.81 tok/sEstimated Auto-generated benchmark | 3GB |
| IlyaGusev/saiga_llama3_8b | Q4 | 744.13 tok/sEstimated Auto-generated benchmark | 4GB |
| HuggingFaceH4/zephyr-7b-beta | Q4 | 743.73 tok/sEstimated Auto-generated benchmark | 4GB |
| tencent/HunyuanVideo-1.5 | Q4 | 742.20 tok/sEstimated Auto-generated benchmark | 4GB |
| Qwen/Qwen2.5-1.5B-Instruct | Q4 | 740.45 tok/sEstimated Auto-generated benchmark | 3GB |
| parler-tts/parler-tts-large-v1 | Q4 | 739.57 tok/sEstimated Auto-generated benchmark | 4GB |
| Qwen/Qwen2.5-7B-Instruct | Q4 | 739.04 tok/sEstimated Auto-generated benchmark | 4GB |
| google/gemma-3-270m-it | Q4 | 736.45 tok/sEstimated Auto-generated benchmark | 4GB |
| Alibaba-NLP/gte-Qwen2-1.5B-instruct | Q4 | 733.49 tok/sEstimated Auto-generated benchmark | 3GB |
| mistralai/Mistral-7B-Instruct-v0.1 | Q4 | 733.07 tok/sEstimated Auto-generated benchmark | 4GB |
| meta-llama/Llama-2-7b-chat-hf | Q4 | 732.36 tok/sEstimated Auto-generated benchmark | 4GB |
| microsoft/phi-4 | Q4 | 731.80 tok/sEstimated Auto-generated benchmark | 4GB |
| dicta-il/dictalm2.0-instruct | Q4 | 731.30 tok/sEstimated Auto-generated benchmark | 4GB |
| Qwen/Qwen2.5-7B-Instruct | Q4 | 729.69 tok/sEstimated Auto-generated benchmark | 4GB |
| meta-llama/Llama-3.1-8B | Q4 | 729.12 tok/sEstimated Auto-generated benchmark | 4GB |
| HuggingFaceTB/SmolLM-135M | Q4 | 727.51 tok/sEstimated Auto-generated benchmark | 4GB |
| lmstudio-community/Qwen3-4B-Thinking-2507-MLX-4bit | Q4 | 727.22 tok/sEstimated Auto-generated benchmark | 2GB |
| meta-llama/Meta-Llama-3-8B-Instruct | Q4 | 724.98 tok/sEstimated Auto-generated benchmark | 4GB |
| lmsys/vicuna-7b-v1.5 | Q4 | 724.21 tok/sEstimated Auto-generated benchmark | 4GB |
| facebook/opt-125m | Q4 | 722.97 tok/sEstimated Auto-generated benchmark | 4GB |
| HuggingFaceTB/SmolLM2-135M | Q4 | 722.22 tok/sEstimated Auto-generated benchmark | 4GB |
| Qwen/Qwen3-8B-Base | Q4 | 721.15 tok/sEstimated Auto-generated benchmark | 4GB |
| deepseek-ai/DeepSeek-R1-0528 | Q4 | 721.03 tok/sEstimated Auto-generated benchmark | 4GB |
| numind/NuExtract-1.5 | Q4 | 720.16 tok/sEstimated Auto-generated benchmark | 4GB |
| meta-llama/Llama-Guard-3-8B | Q4 | 718.74 tok/sEstimated Auto-generated benchmark | 4GB |
| Qwen/Qwen2.5-7B | Q4 | 717.28 tok/sEstimated Auto-generated benchmark | 4GB |
| skt/kogpt2-base-v2 | Q4 | 715.47 tok/sEstimated Auto-generated benchmark | 4GB |
| EleutherAI/gpt-neo-125m | Q4 | 713.82 tok/sEstimated Auto-generated benchmark | 4GB |
| Qwen/Qwen3-4B | Q4 | 710.79 tok/sEstimated Auto-generated benchmark | 2GB |
| deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct | Q4 | 709.99 tok/sEstimated Auto-generated benchmark | 4GB |
| mistralai/Mistral-7B-v0.1 | Q4 | 705.00 tok/sEstimated Auto-generated benchmark | 4GB |
| GSAI-ML/LLaDA-8B-Instruct | Q4 | 704.54 tok/sEstimated Auto-generated benchmark | 4GB |
| llamafactory/tiny-random-Llama-3 | Q4 | 704.02 tok/sEstimated Auto-generated benchmark | 4GB |
| Qwen/Qwen2.5-Math-1.5B | Q4 | 703.78 tok/sEstimated Auto-generated benchmark | 3GB |
| microsoft/DialoGPT-medium | Q4 | 703.01 tok/sEstimated Auto-generated benchmark | 4GB |
| Qwen/Qwen2.5-1.5B | Q4 | 701.71 tok/sEstimated Auto-generated benchmark | 3GB |
| BSC-LT/salamandraTA-7b-instruct | Q4 | 700.10 tok/sEstimated Auto-generated benchmark | 4GB |
| microsoft/Phi-3.5-vision-instruct | Q4 | 697.62 tok/sEstimated Auto-generated benchmark | 4GB |
| sshleifer/tiny-gpt2 | Q4 | 694.61 tok/sEstimated Auto-generated benchmark | 4GB |
| bigscience/bloomz-560m | Q4 | 694.58 tok/sEstimated Auto-generated benchmark | 4GB |
| microsoft/Phi-4-mini-instruct | Q4 | 694.02 tok/sEstimated Auto-generated benchmark | 4GB |
| unsloth/mistral-7b-v0.3-bnb-4bit | Q4 | 693.08 tok/sEstimated Auto-generated benchmark | 4GB |
| EleutherAI/pythia-70m-deduped | Q4 | 692.68 tok/sEstimated Auto-generated benchmark | 4GB |
| microsoft/Phi-3.5-mini-instruct | Q4 | 691.94 tok/sEstimated Auto-generated benchmark | 4GB |
| openai-community/gpt2-medium | Q4 | 691.79 tok/sEstimated Auto-generated benchmark | 4GB |
| microsoft/DialoGPT-small | Q4 | 691.36 tok/sEstimated Auto-generated benchmark | 4GB |
| deepseek-ai/DeepSeek-R1-Distill-Qwen-7B | Q4 | 691.29 tok/sEstimated Auto-generated benchmark | 4GB |
| kaitchup/Phi-3-mini-4k-instruct-gptq-4bit | Q4 | 687.23 tok/sEstimated Auto-generated benchmark | 2GB |
| microsoft/VibeVoice-1.5B | Q4 | 686.19 tok/sEstimated Auto-generated benchmark | 3GB |
| Qwen/Qwen3-8B-FP8 | Q4 | 685.29 tok/sEstimated Auto-generated benchmark | 4GB |
| Qwen/Qwen2-1.5B-Instruct | Q4 | 684.83 tok/sEstimated Auto-generated benchmark | 3GB |
| hmellor/tiny-random-LlamaForCausalLM | Q4 | 684.61 tok/sEstimated Auto-generated benchmark | 4GB |
| ibm-granite/granite-3.3-8b-instruct | Q4 | 682.53 tok/sEstimated Auto-generated benchmark | 4GB |
| deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B | Q4 | 680.96 tok/sEstimated Auto-generated benchmark | 3GB |
| microsoft/Phi-3-mini-4k-instruct | Q4 | 679.77 tok/sEstimated Auto-generated benchmark | 4GB |
| black-forest-labs/FLUX.1-dev | Q4 | 676.48 tok/sEstimated Auto-generated benchmark | 4GB |
| Qwen/Qwen3-1.7B | Q4 | 676.33 tok/sEstimated Auto-generated benchmark | 4GB |
| meta-llama/Llama-3.2-3B-Instruct | Q4 | 675.18 tok/sEstimated Auto-generated benchmark | 2GB |
| GSAI-ML/LLaDA-8B-Base | Q4 | 674.38 tok/sEstimated Auto-generated benchmark | 4GB |
| vikhyatk/moondream2 | Q4 | 673.06 tok/sEstimated Auto-generated benchmark | 4GB |
| lmstudio-community/DeepSeek-R1-0528-Qwen3-8B-MLX-8bit | Q4 | 672.29 tok/sEstimated Auto-generated benchmark | 4GB |
| microsoft/Phi-3-mini-128k-instruct | Q4 | 671.72 tok/sEstimated Auto-generated benchmark | 4GB |
| microsoft/phi-2 | Q4 | 668.90 tok/sEstimated Auto-generated benchmark | 4GB |
| deepseek-ai/DeepSeek-V3-0324 | Q4 | 667.43 tok/sEstimated Auto-generated benchmark | 4GB |
| Qwen/Qwen3-4B-Thinking-2507-FP8 | Q4 | 664.92 tok/sEstimated Auto-generated benchmark | 2GB |
| openai-community/gpt2-xl | Q4 | 664.40 tok/sEstimated Auto-generated benchmark | 4GB |
| liuhaotian/llava-v1.5-7b | Q4 | 663.30 tok/sEstimated Auto-generated benchmark | 4GB |
| Qwen/Qwen3-0.6B | Q4 | 662.43 tok/sEstimated Auto-generated benchmark | 3GB |
| deepseek-ai/DeepSeek-V3 | Q4 | 661.09 tok/sEstimated Auto-generated benchmark | 4GB |
| trl-internal-testing/tiny-LlamaForCausalLM-3.2 | Q4 | 660.47 tok/sEstimated Auto-generated benchmark | 4GB |
| Qwen/Qwen3-Embedding-8B | Q4 | 659.59 tok/sEstimated Auto-generated benchmark | 4GB |
| deepseek-ai/DeepSeek-R1 | Q4 | 657.58 tok/sEstimated Auto-generated benchmark | 4GB |
| unsloth/Meta-Llama-3.1-8B-Instruct | Q4 | 656.74 tok/sEstimated Auto-generated benchmark | 4GB |
| meta-llama/Llama-3.1-8B-Instruct | Q4 | 656.27 tok/sEstimated Auto-generated benchmark | 4GB |
| Qwen/Qwen2-0.5B-Instruct | Q4 | 655.15 tok/sEstimated Auto-generated benchmark | 3GB |
| openai-community/gpt2 | Q4 | 654.57 tok/sEstimated Auto-generated benchmark | 4GB |
| lmstudio-community/DeepSeek-R1-0528-Qwen3-8B-MLX-4bit | Q4 | 654.14 tok/sEstimated Auto-generated benchmark | 4GB |
| microsoft/Phi-3.5-mini-instruct | Q4 | 654.01 tok/sEstimated Auto-generated benchmark | 2GB |
| lmstudio-community/Qwen3-4B-Thinking-2507-MLX-8bit | Q4 | 653.89 tok/sEstimated Auto-generated benchmark | 2GB |
| rinna/japanese-gpt-neox-small | Q4 | 653.44 tok/sEstimated Auto-generated benchmark | 4GB |
| Qwen/Qwen3-Embedding-0.6B | Q4 | 653.23 tok/sEstimated Auto-generated benchmark | 3GB |
| zai-org/GLM-4.6-FP8 | Q4 | 652.31 tok/sEstimated Auto-generated benchmark | 4GB |
| NousResearch/Meta-Llama-3.1-8B-Instruct | Q4 | 650.83 tok/sEstimated Auto-generated benchmark | 4GB |
| allenai/Olmo-3-7B-Think | Q4 | 649.74 tok/sEstimated Auto-generated benchmark | 4GB |
| black-forest-labs/FLUX.2-dev | Q4 | 649.42 tok/sEstimated Auto-generated benchmark | 4GB |
| rednote-hilab/dots.ocr | Q4 | 648.47 tok/sEstimated Auto-generated benchmark | 4GB |
| Qwen/Qwen3-Reranker-0.6B | Q4 | 648.18 tok/sEstimated Auto-generated benchmark | 3GB |
| Qwen/Qwen3-1.7B-Base | Q4 | 647.68 tok/sEstimated Auto-generated benchmark | 4GB |
| Qwen/Qwen2-0.5B | Q4 | 646.89 tok/sEstimated Auto-generated benchmark | 3GB |
| HuggingFaceM4/tiny-random-LlamaForCausalLM | Q4 | 645.69 tok/sEstimated Auto-generated benchmark | 4GB |
| openai-community/gpt2-large | Q4 | 643.83 tok/sEstimated Auto-generated benchmark | 4GB |
| Qwen/Qwen3-8B | Q4 | 643.77 tok/sEstimated Auto-generated benchmark | 4GB |
| zai-org/GLM-4.5-Air | Q4 | 643.38 tok/sEstimated Auto-generated benchmark | 4GB |
| mistralai/Mistral-7B-Instruct-v0.2 | Q4 | 642.93 tok/sEstimated Auto-generated benchmark | 4GB |
| Tongyi-MAI/Z-Image-Turbo | Q4 | 640.26 tok/sEstimated Auto-generated benchmark | 4GB |
| MiniMaxAI/MiniMax-M2 | Q4 | 639.35 tok/sEstimated Auto-generated benchmark | 4GB |
| Qwen/Qwen3-Embedding-4B | Q4 | 637.56 tok/sEstimated Auto-generated benchmark | 2GB |
| Qwen/Qwen3-4B-Base | Q4 | 637.39 tok/sEstimated Auto-generated benchmark | 2GB |
| Gensyn/Qwen2.5-0.5B-Instruct | Q4 | 633.35 tok/sEstimated Auto-generated benchmark | 3GB |
| google-t5/t5-3b | Q8 | 633.16 tok/sEstimated Auto-generated benchmark | 3GB |
| distilbert/distilgpt2 | Q4 | 631.48 tok/sEstimated Auto-generated benchmark | 4GB |
| unsloth/Llama-3.2-3B-Instruct | Q8 | 631.17 tok/sEstimated Auto-generated benchmark | 3GB |
| Qwen/Qwen3-4B-Thinking-2507 | Q4 | 630.67 tok/sEstimated Auto-generated benchmark | 2GB |
| Qwen/Qwen2.5-0.5B-Instruct | Q4 | 629.27 tok/sEstimated Auto-generated benchmark | 3GB |
| unsloth/Llama-3.2-1B-Instruct | Q8 | 628.59 tok/sEstimated Auto-generated benchmark | 1GB |
| unsloth/Meta-Llama-3.1-8B-Instruct-bnb-4bit | Q4 | 627.76 tok/sEstimated Auto-generated benchmark | 4GB |
| context-labs/meta-llama-Llama-3.2-3B-Instruct-FP16 | Q8 | 627.21 tok/sEstimated Auto-generated benchmark | 3GB |
| meta-llama/Llama-3.2-3B-Instruct | Q8 | 627.03 tok/sEstimated Auto-generated benchmark | 3GB |
| trl-internal-testing/tiny-random-LlamaForCausalLM | Q4 | 626.15 tok/sEstimated Auto-generated benchmark | 4GB |
| meta-llama/Meta-Llama-3-8B | Q4 | 625.53 tok/sEstimated Auto-generated benchmark | 4GB |
| ibm-granite/granite-docling-258M | Q4 | 621.13 tok/sEstimated Auto-generated benchmark | 4GB |
| Qwen/Qwen2-7B-Instruct | Q4 | 620.89 tok/sEstimated Auto-generated benchmark | 4GB |
| Qwen/Qwen2.5-0.5B | Q4 | 620.64 tok/sEstimated Auto-generated benchmark | 3GB |
| deepseek-ai/DeepSeek-V3.1 | Q4 | 620.51 tok/sEstimated Auto-generated benchmark | 4GB |
| TinyLlama/TinyLlama-1.1B-Chat-v1.0 | Q8 | 612.43 tok/sEstimated Auto-generated benchmark | 1GB |
| WeiboAI/VibeThinker-1.5B | Q8 | 608.69 tok/sEstimated Auto-generated benchmark | 2GB |
| apple/OpenELM-1_1B-Instruct | Q8 | 597.43 tok/sEstimated Auto-generated benchmark | 1GB |
| unsloth/gemma-3-1b-it | Q8 | 596.97 tok/sEstimated Auto-generated benchmark | 1GB |
| Qwen/Qwen2.5-3B-Instruct | Q8 | 596.38 tok/sEstimated Auto-generated benchmark | 3GB |
| meta-llama/Llama-Guard-3-1B | Q8 | 589.46 tok/sEstimated Auto-generated benchmark | 1GB |
| google/embeddinggemma-300m | Q8 | 585.61 tok/sEstimated Auto-generated benchmark | 1GB |
| facebook/sam3 | Q8 | 584.50 tok/sEstimated Auto-generated benchmark | 1GB |
| google-bert/bert-base-uncased | Q8 | 580.24 tok/sEstimated Auto-generated benchmark | 1GB |
| LiquidAI/LFM2-1.2B | Q8 | 579.15 tok/sEstimated Auto-generated benchmark | 2GB |
| Qwen/Qwen2.5-3B | Q8 | 578.68 tok/sEstimated Auto-generated benchmark | 3GB |
| ibm-research/PowerMoE-3b | Q8 | 571.20 tok/sEstimated Auto-generated benchmark | 3GB |
| inference-net/Schematron-3B | Q8 | 565.33 tok/sEstimated Auto-generated benchmark | 3GB |
| tencent/HunyuanOCR | Q8 | 563.83 tok/sEstimated Auto-generated benchmark | 2GB |
| meta-llama/Llama-3.1-8B-Instruct | Q4 | 562.55 tok/sEstimated Auto-generated benchmark | 4GB |
| bigcode/starcoder2-3b | Q8 | 561.60 tok/sEstimated Auto-generated benchmark | 3GB |
| meta-llama/Llama-3.2-3B | Q8 | 561.04 tok/sEstimated Auto-generated benchmark | 3GB |
| allenai/OLMo-2-0425-1B | Q8 | 558.44 tok/sEstimated Auto-generated benchmark | 1GB |
| google/gemma-2-9b-it | Q4 | 554.32 tok/sEstimated Auto-generated benchmark | 5GB |
| ibm-granite/granite-3.3-2b-instruct | Q8 | 552.07 tok/sEstimated Auto-generated benchmark | 2GB |
| deepseek-ai/DeepSeek-OCR | Q8 | 549.96 tok/sEstimated Auto-generated benchmark | 4GB |
| Qwen/Qwen3-14B-Base | Q4 | 546.51 tok/sEstimated Auto-generated benchmark | 7GB |
| google/gemma-2-2b-it | Q8 | 545.02 tok/sEstimated Auto-generated benchmark | 2GB |
| nari-labs/Dia2-2B | Q8 | 545.00 tok/sEstimated Auto-generated benchmark | 3GB |
| google/gemma-3-1b-it | Q8 | 542.85 tok/sEstimated Auto-generated benchmark | 1GB |
| deepseek-ai/deepseek-coder-1.3b-instruct | Q8 | 535.56 tok/sEstimated Auto-generated benchmark | 3GB |
| meta-llama/Llama-3.2-1B-Instruct | Q8 | 534.88 tok/sEstimated Auto-generated benchmark | 1GB |
| Qwen/Qwen2.5-14B-Instruct | Q4 | 532.29 tok/sEstimated Auto-generated benchmark | 7GB |
| meta-llama/Llama-2-7b-chat-hf | Q8 | 529.63 tok/sEstimated Auto-generated benchmark | 7GB |
| Qwen/Qwen-Image-Edit-2509 | Q8 | 529.52 tok/sEstimated Auto-generated benchmark | 8GB |
| microsoft/Phi-3-mini-128k-instruct | Q8 | 529.49 tok/sEstimated Auto-generated benchmark | 7GB |
| lmstudio-community/Qwen3-4B-Thinking-2507-MLX-6bit | Q8 | 528.99 tok/sEstimated Auto-generated benchmark | 4GB |
| Qwen/Qwen3-0.6B | Q8 | 528.95 tok/sEstimated Auto-generated benchmark | 6GB |
| meta-llama/Llama-3.2-1B | Q8 | 527.73 tok/sEstimated Auto-generated benchmark | 1GB |
| huggyllama/llama-7b | Q8 | 527.69 tok/sEstimated Auto-generated benchmark | 7GB |
| microsoft/Phi-4-mini-instruct | Q8 | 527.27 tok/sEstimated Auto-generated benchmark | 7GB |
| Qwen/Qwen3-Embedding-4B | Q8 | 527.13 tok/sEstimated Auto-generated benchmark | 4GB |
| microsoft/Phi-3-mini-4k-instruct | Q8 | 525.93 tok/sEstimated Auto-generated benchmark | 7GB |
| Qwen/Qwen3-4B-Thinking-2507-FP8 | Q8 | 525.68 tok/sEstimated Auto-generated benchmark | 4GB |
| openai-community/gpt2 | Q8 | 525.37 tok/sEstimated Auto-generated benchmark | 7GB |
| numind/NuExtract-1.5 | Q8 | 525.27 tok/sEstimated Auto-generated benchmark | 7GB |
| Qwen/Qwen3-Embedding-0.6B | Q8 | 525.00 tok/sEstimated Auto-generated benchmark | 6GB |
| google/gemma-2b | Q8 | 523.24 tok/sEstimated Auto-generated benchmark | 2GB |
| Qwen/Qwen2.5-14B | Q4 | 522.94 tok/sEstimated Auto-generated benchmark | 7GB |
| Qwen/Qwen2.5-7B | Q8 | 521.84 tok/sEstimated Auto-generated benchmark | 7GB |
| lmsys/vicuna-7b-v1.5 | Q8 | 520.52 tok/sEstimated Auto-generated benchmark | 7GB |
| distilbert/distilgpt2 | Q8 | 520.52 tok/sEstimated Auto-generated benchmark | 7GB |
| ibm-granite/granite-docling-258M | Q8 | 519.68 tok/sEstimated Auto-generated benchmark | 7GB |
| zai-org/GLM-4.6-FP8 | Q8 | 518.98 tok/sEstimated Auto-generated benchmark | 7GB |
| IlyaGusev/saiga_llama3_8b | Q8 | 517.89 tok/sEstimated Auto-generated benchmark | 9GB |
| trl-internal-testing/tiny-random-LlamaForCausalLM | Q8 | 516.62 tok/sEstimated Auto-generated benchmark | 7GB |
| MiniMaxAI/MiniMax-M2 | Q8 | 515.70 tok/sEstimated Auto-generated benchmark | 7GB |
| deepseek-ai/DeepSeek-V3.1 | Q8 | 512.79 tok/sEstimated Auto-generated benchmark | 7GB |
| Qwen/Qwen3-1.7B | Q8 | 512.45 tok/sEstimated Auto-generated benchmark | 7GB |
| meta-llama/Llama-2-7b-hf | Q8 | 510.14 tok/sEstimated Auto-generated benchmark | 7GB |
| unsloth/Meta-Llama-3.1-8B-Instruct-bnb-4bit | Q8 | 509.91 tok/sEstimated Auto-generated benchmark | 9GB |
| meta-llama/Meta-Llama-3-8B-Instruct | Q8 | 509.42 tok/sEstimated Auto-generated benchmark | 9GB |
| GSAI-ML/LLaDA-8B-Base | Q8 | 509.15 tok/sEstimated Auto-generated benchmark | 9GB |
| Qwen/Qwen3-4B-Base | Q8 | 508.92 tok/sEstimated Auto-generated benchmark | 4GB |
| ibm-granite/granite-3.3-8b-instruct | Q8 | 505.86 tok/sEstimated Auto-generated benchmark | 9GB |
| Qwen/Qwen2.5-Coder-1.5B | Q8 | 505.71 tok/sEstimated Auto-generated benchmark | 5GB |
| liuhaotian/llava-v1.5-7b | Q8 | 505.56 tok/sEstimated Auto-generated benchmark | 7GB |
| Qwen/Qwen3-8B-Base | Q8 | 503.98 tok/sEstimated Auto-generated benchmark | 9GB |
| lmstudio-community/Qwen3-4B-Thinking-2507-MLX-4bit | Q8 | 502.16 tok/sEstimated Auto-generated benchmark | 4GB |
| Qwen/Qwen2.5-7B-Instruct | Q8 | 499.81 tok/sEstimated Auto-generated benchmark | 7GB |
| Qwen/Qwen3-8B | Q8 | 498.75 tok/sEstimated Auto-generated benchmark | 9GB |
| mistralai/Ministral-3-14B-Instruct-2512 | Q4 | 498.23 tok/sEstimated Auto-generated benchmark | 8GB |
| kaitchup/Phi-3-mini-4k-instruct-gptq-4bit | Q8 | 496.81 tok/sEstimated Auto-generated benchmark | 4GB |
| Gensyn/Qwen2.5-0.5B-Instruct | Q8 | 496.74 tok/sEstimated Auto-generated benchmark | 5GB |
| NousResearch/Meta-Llama-3.1-8B-Instruct | Q8 | 496.55 tok/sEstimated Auto-generated benchmark | 9GB |
| skt/kogpt2-base-v2 | Q8 | 496.54 tok/sEstimated Auto-generated benchmark | 7GB |
| microsoft/phi-4 | Q8 | 496.45 tok/sEstimated Auto-generated benchmark | 7GB |
| microsoft/Phi-3-medium-128k-instruct | Q4 | 496.24 tok/sEstimated Auto-generated benchmark | 7GB |
| Tongyi-MAI/Z-Image-Turbo | Q8 | 495.85 tok/sEstimated Auto-generated benchmark | 8GB |
| unsloth/Meta-Llama-3.1-8B-Instruct | Q8 | 495.83 tok/sEstimated Auto-generated benchmark | 9GB |
| unsloth/mistral-7b-v0.3-bnb-4bit | Q8 | 495.80 tok/sEstimated Auto-generated benchmark | 7GB |
| BSC-LT/salamandraTA-7b-instruct | Q8 | 495.39 tok/sEstimated Auto-generated benchmark | 7GB |
| deepseek-ai/DeepSeek-V3-0324 | Q8 | 494.66 tok/sEstimated Auto-generated benchmark | 7GB |
| meta-llama/Llama-Guard-3-8B | Q8 | 494.61 tok/sEstimated Auto-generated benchmark | 9GB |
| nvidia/NVIDIA-Nemotron-Nano-9B-v2 | Q4 | 493.21 tok/sEstimated Auto-generated benchmark | 5GB |
| lmstudio-community/DeepSeek-R1-0528-Qwen3-8B-MLX-8bit | Q8 | 493.05 tok/sEstimated Auto-generated benchmark | 9GB |
| Alibaba-NLP/gte-Qwen2-1.5B-instruct | Q8 | 490.90 tok/sEstimated Auto-generated benchmark | 5GB |
| openai-community/gpt2-medium | Q8 | 489.97 tok/sEstimated Auto-generated benchmark | 7GB |
| EssentialAI/rnj-1 | Q4 | 487.85 tok/sEstimated Auto-generated benchmark | 5GB |
| EleutherAI/gpt-neo-125m | Q8 | 486.78 tok/sEstimated Auto-generated benchmark | 7GB |
| Qwen/Qwen2.5-14B-Instruct | Q4 | 486.56 tok/sEstimated Auto-generated benchmark | 8GB |
| microsoft/Phi-3.5-vision-instruct | Q8 | 486.24 tok/sEstimated Auto-generated benchmark | 7GB |
| lmstudio-community/Qwen3-4B-Thinking-2507-MLX-8bit | Q8 | 486.05 tok/sEstimated Auto-generated benchmark | 4GB |
| meta-llama/Llama-2-13b-chat-hf | Q4 | 485.81 tok/sEstimated Auto-generated benchmark | 7GB |
| Qwen/Qwen3-4B-Instruct-2507 | Q8 | 483.30 tok/sEstimated Auto-generated benchmark | 4GB |
| ai-forever/ruGPT-3.5-13B | Q4 | 482.72 tok/sEstimated Auto-generated benchmark | 7GB |
| sshleifer/tiny-gpt2 | Q8 | 482.01 tok/sEstimated Auto-generated benchmark | 7GB |
| Qwen/Qwen3-8B-FP8 | Q8 | 480.63 tok/sEstimated Auto-generated benchmark | 9GB |
| tencent/HunyuanVideo-1.5 | Q8 | 479.89 tok/sEstimated Auto-generated benchmark | 8GB |
| microsoft/VibeVoice-1.5B | Q8 | 479.50 tok/sEstimated Auto-generated benchmark | 5GB |
| microsoft/DialoGPT-medium | Q8 | 479.22 tok/sEstimated Auto-generated benchmark | 7GB |
| hmellor/tiny-random-LlamaForCausalLM | Q8 | 477.78 tok/sEstimated Auto-generated benchmark | 7GB |
| microsoft/DialoGPT-small | Q8 | 477.74 tok/sEstimated Auto-generated benchmark | 7GB |
| rednote-hilab/dots.ocr | Q8 | 476.06 tok/sEstimated Auto-generated benchmark | 7GB |
| zai-org/GLM-4.5-Air | Q8 | 475.77 tok/sEstimated Auto-generated benchmark | 7GB |
| Qwen/Qwen3-Reranker-0.6B | Q8 | 474.94 tok/sEstimated Auto-generated benchmark | 6GB |
| google/gemma-3-270m-it | Q8 | 474.50 tok/sEstimated Auto-generated benchmark | 7GB |
| microsoft/Phi-3.5-mini-instruct | Q8 | 474.39 tok/sEstimated Auto-generated benchmark | 4GB |
| deepseek-ai/DeepSeek-V3 | Q8 | 474.17 tok/sEstimated Auto-generated benchmark | 7GB |
| parler-tts/parler-tts-large-v1 | Q8 | 473.80 tok/sEstimated Auto-generated benchmark | 7GB |
| rinna/japanese-gpt-neox-small | Q8 | 472.83 tok/sEstimated Auto-generated benchmark | 7GB |
| Qwen/Qwen2-1.5B-Instruct | Q8 | 472.76 tok/sEstimated Auto-generated benchmark | 5GB |
| Qwen/Qwen3-14B | Q4 | 472.20 tok/sEstimated Auto-generated benchmark | 7GB |
| dicta-il/dictalm2.0-instruct | Q8 | 471.20 tok/sEstimated Auto-generated benchmark | 7GB |
| OpenPipe/Qwen3-14B-Instruct | Q4 | 470.96 tok/sEstimated Auto-generated benchmark | 7GB |
| Qwen/Qwen3-0.6B-Base | Q8 | 469.93 tok/sEstimated Auto-generated benchmark | 6GB |
| deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct | Q8 | 468.37 tok/sEstimated Auto-generated benchmark | 7GB |
| NousResearch/Hermes-3-Llama-3.1-8B | Q4 | 467.95 tok/sEstimated Auto-generated benchmark | 4GB |
| HuggingFaceM4/tiny-random-LlamaForCausalLM | Q8 | 467.02 tok/sEstimated Auto-generated benchmark | 7GB |
| openai-community/gpt2-large | Q8 | 465.78 tok/sEstimated Auto-generated benchmark | 7GB |
| mistralai/Mistral-7B-Instruct-v0.2 | Q8 | 465.18 tok/sEstimated Auto-generated benchmark | 7GB |
| Qwen/Qwen2.5-0.5B | Q8 | 463.81 tok/sEstimated Auto-generated benchmark | 5GB |
| meta-llama/Llama-3.1-8B-Instruct | Q8 | 462.72 tok/sEstimated Auto-generated benchmark | 9GB |
| microsoft/Phi-4-multimodal-instruct | Q8 | 462.09 tok/sEstimated Auto-generated benchmark | 7GB |
| bigscience/bloomz-560m | Q8 | 462.01 tok/sEstimated Auto-generated benchmark | 7GB |
| GSAI-ML/LLaDA-8B-Instruct | Q8 | 461.86 tok/sEstimated Auto-generated benchmark | 9GB |
| Qwen/Qwen3-4B | Q8 | 461.81 tok/sEstimated Auto-generated benchmark | 4GB |
| llamafactory/tiny-random-Llama-3 | Q8 | 460.34 tok/sEstimated Auto-generated benchmark | 7GB |
| openai-community/gpt2-xl | Q8 | 460.22 tok/sEstimated Auto-generated benchmark | 7GB |
| meta-llama/Llama-3.2-3B-Instruct | Q8 | 458.16 tok/sEstimated Auto-generated benchmark | 3GB |
| black-forest-labs/FLUX.2-dev | Q8 | 457.65 tok/sEstimated Auto-generated benchmark | 8GB |
| petals-team/StableBeluga2 | Q8 | 457.19 tok/sEstimated Auto-generated benchmark | 7GB |
| Qwen/Qwen2.5-0.5B-Instruct | Q8 | 457.15 tok/sEstimated Auto-generated benchmark | 5GB |
| Qwen/Qwen2.5-7B-Instruct | Q8 | 456.26 tok/sEstimated Auto-generated benchmark | 8GB |
| trl-internal-testing/tiny-Qwen2ForCausalLM-2.5 | Q8 | 454.49 tok/sEstimated Auto-generated benchmark | 7GB |
| deepseek-ai/DeepSeek-R1-Distill-Qwen-7B | Q8 | 453.74 tok/sEstimated Auto-generated benchmark | 7GB |
| trl-internal-testing/tiny-LlamaForCausalLM-3.2 | Q8 | 453.55 tok/sEstimated Auto-generated benchmark | 7GB |
| microsoft/Phi-3.5-mini-instruct | Q8 | 450.03 tok/sEstimated Auto-generated benchmark | 7GB |
| mistralai/Mistral-7B-Instruct-v0.1 | Q8 | 449.38 tok/sEstimated Auto-generated benchmark | 7GB |
| HuggingFaceH4/zephyr-7b-beta | Q8 | 448.66 tok/sEstimated Auto-generated benchmark | 7GB |
| Qwen/Qwen3-4B-Thinking-2507 | Q8 | 447.75 tok/sEstimated Auto-generated benchmark | 4GB |
| Qwen/Qwen2.5-1.5B-Instruct | Q8 | 447.65 tok/sEstimated Auto-generated benchmark | 5GB |
| Qwen/Qwen2-0.5B-Instruct | Q8 | 446.55 tok/sEstimated Auto-generated benchmark | 5GB |
| Qwen/Qwen2.5-Math-1.5B | Q8 | 446.50 tok/sEstimated Auto-generated benchmark | 5GB |
| mistralai/Mistral-7B-v0.1 | Q8 | 446.29 tok/sEstimated Auto-generated benchmark | 7GB |
| lmstudio-community/DeepSeek-R1-0528-Qwen3-8B-MLX-4bit | Q8 | 446.22 tok/sEstimated Auto-generated benchmark | 9GB |
| deepseek-ai/DeepSeek-R1-Distill-Llama-8B | Q8 | 446.12 tok/sEstimated Auto-generated benchmark | 9GB |
| Qwen/Qwen3-Embedding-8B | Q8 | 445.59 tok/sEstimated Auto-generated benchmark | 9GB |
| HuggingFaceTB/SmolLM2-135M | Q8 | 445.39 tok/sEstimated Auto-generated benchmark | 7GB |
| deepseek-ai/DeepSeek-R1-0528 | Q8 | 445.02 tok/sEstimated Auto-generated benchmark | 7GB |
| Qwen/Qwen2-0.5B | Q8 | 444.69 tok/sEstimated Auto-generated benchmark | 5GB |
| vikhyatk/moondream2 | Q8 | 444.12 tok/sEstimated Auto-generated benchmark | 7GB |
| meta-llama/Llama-3.1-8B | Q8 | 443.98 tok/sEstimated Auto-generated benchmark | 9GB |
| EleutherAI/pythia-70m-deduped | Q8 | 443.11 tok/sEstimated Auto-generated benchmark | 7GB |
| allenai/Olmo-3-7B-Think | Q8 | 443.05 tok/sEstimated Auto-generated benchmark | 8GB |
| facebook/opt-125m | Q8 | 442.72 tok/sEstimated Auto-generated benchmark | 7GB |
| meta-llama/Meta-Llama-3-8B | Q8 | 442.60 tok/sEstimated Auto-generated benchmark | 9GB |
| deepseek-ai/DeepSeek-R1 | Q8 | 442.24 tok/sEstimated Auto-generated benchmark | 7GB |
| black-forest-labs/FLUX.1-dev | Q8 | 440.97 tok/sEstimated Auto-generated benchmark | 8GB |
| deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B | Q8 | 440.85 tok/sEstimated Auto-generated benchmark | 5GB |
| swiss-ai/Apertus-8B-Instruct-2509 | Q8 | 440.42 tok/sEstimated Auto-generated benchmark | 9GB |
| microsoft/phi-2 | Q8 | 438.76 tok/sEstimated Auto-generated benchmark | 7GB |
| Qwen/Qwen2-7B-Instruct | Q8 | 437.92 tok/sEstimated Auto-generated benchmark | 7GB |
| Qwen/Qwen2.5-1.5B | Q8 | 437.18 tok/sEstimated Auto-generated benchmark | 5GB |
| HuggingFaceTB/SmolLM-135M | Q8 | 437.16 tok/sEstimated Auto-generated benchmark | 7GB |
| Qwen/Qwen3-1.7B-Base | Q8 | 436.72 tok/sEstimated Auto-generated benchmark | 7GB |
| Qwen/Qwen2.5-Coder-7B-Instruct | Q8 | 436.17 tok/sEstimated Auto-generated benchmark | 7GB |
| unsloth/gpt-oss-20b-unsloth-bnb-4bit | Q4 | 409.56 tok/sEstimated Auto-generated benchmark | 10GB |
| openai/gpt-oss-safeguard-20b | Q4 | 407.36 tok/sEstimated Auto-generated benchmark | 11GB |
| lmstudio-community/Qwen3-Coder-30B-A3B-Instruct-MLX-4bit | Q4 | 402.27 tok/sEstimated Auto-generated benchmark | 15GB |
| Qwen/Qwen3-30B-A3B-Thinking-2507 | Q4 | 401.16 tok/sEstimated Auto-generated benchmark | 15GB |
| mistralai/Mistral-Small-Instruct-2409 | Q4 | 397.86 tok/sEstimated Auto-generated benchmark | 11GB |
| NousResearch/Hermes-3-Llama-3.1-8B | Q8 | 395.23 tok/sEstimated Auto-generated benchmark | 9GB |
| lmstudio-community/Qwen3-Coder-30B-A3B-Instruct-MLX-6bit | Q4 | 388.69 tok/sEstimated Auto-generated benchmark | 15GB |
| openai/gpt-oss-20b | Q4 | 388.61 tok/sEstimated Auto-generated benchmark | 10GB |
| mistralai/Ministral-3-14B-Instruct-2512 | Q8 | 386.57 tok/sEstimated Auto-generated benchmark | 16GB |
| mlx-community/gpt-oss-20b-MXFP4-Q8 | Q4 | 384.15 tok/sEstimated Auto-generated benchmark | 10GB |
| Qwen/Qwen2.5-14B-Instruct | Q8 | 378.95 tok/sEstimated Auto-generated benchmark | 14GB |
| google/gemma-2-27b-it | Q4 | 377.43 tok/sEstimated Auto-generated benchmark | 14GB |
| google/gemma-2-9b-it | Q8 | 375.56 tok/sEstimated Auto-generated benchmark | 10GB |
| Qwen/Qwen3-30B-A3B | Q4 | 371.30 tok/sEstimated Auto-generated benchmark | 15GB |
| unsloth/gpt-oss-20b-BF16 | Q4 | 369.14 tok/sEstimated Auto-generated benchmark | 10GB |
| Qwen/Qwen2.5-14B | Q8 | 368.91 tok/sEstimated Auto-generated benchmark | 14GB |
| Qwen/Qwen3-30B-A3B-Instruct-2507 | Q4 | 368.76 tok/sEstimated Auto-generated benchmark | 15GB |
| Qwen/Qwen3-14B-Base | Q8 | 367.52 tok/sEstimated Auto-generated benchmark | 14GB |
| EssentialAI/rnj-1 | Q8 | 365.37 tok/sEstimated Auto-generated benchmark | 10GB |
| Qwen/Qwen3-Coder-30B-A3B-Instruct | Q4 | 364.79 tok/sEstimated Auto-generated benchmark | 15GB |
| Qwen/Qwen3-30B-A3B-Instruct-2507-FP8 | Q4 | 364.68 tok/sEstimated Auto-generated benchmark | 15GB |
| ai-forever/ruGPT-3.5-13B | Q8 | 357.72 tok/sEstimated Auto-generated benchmark | 13GB |
| meta-llama/Llama-3.1-8B-Instruct | Q8 | 352.67 tok/sEstimated Auto-generated benchmark | 9GB |
| Qwen/Qwen2.5-14B-Instruct | Q8 | 350.78 tok/sEstimated Auto-generated benchmark | 15GB |
| nvidia/NVIDIA-Nemotron-Nano-9B-v2 | Q8 | 347.75 tok/sEstimated Auto-generated benchmark | 10GB |
| lmstudio-community/Qwen3-Coder-30B-A3B-Instruct-MLX-8bit | Q4 | 346.15 tok/sEstimated Auto-generated benchmark | 15GB |
| deepseek-ai/deepseek-coder-1.3b-instruct | FP16 | 344.89 tok/sEstimated Auto-generated benchmark | 6GB |
| google-t5/t5-3b | FP16 | 343.58 tok/sEstimated Auto-generated benchmark | 6GB |
| lmstudio-community/Qwen3-Coder-30B-A3B-Instruct-MLX-5bit | Q4 | 342.29 tok/sEstimated Auto-generated benchmark | 15GB |
| meta-llama/Llama-2-13b-chat-hf | Q8 | 341.74 tok/sEstimated Auto-generated benchmark | 13GB |
| WeiboAI/VibeThinker-1.5B | FP16 | 340.56 tok/sEstimated Auto-generated benchmark | 4GB |
| ibm-granite/granite-3.3-2b-instruct | FP16 | 340.43 tok/sEstimated Auto-generated benchmark | 4GB |
| unsloth/Llama-3.2-3B-Instruct | FP16 | 339.01 tok/sEstimated Auto-generated benchmark | 6GB |
| OpenPipe/Qwen3-14B-Instruct | Q8 | 338.80 tok/sEstimated Auto-generated benchmark | 14GB |
| apple/OpenELM-1_1B-Instruct | FP16 | 334.87 tok/sEstimated Auto-generated benchmark | 2GB |
| microsoft/Phi-3-medium-128k-instruct | Q8 | 334.67 tok/sEstimated Auto-generated benchmark | 14GB |
| allenai/OLMo-2-0425-1B | FP16 | 331.91 tok/sEstimated Auto-generated benchmark | 2GB |
| LiquidAI/LFM2-1.2B | FP16 | 329.18 tok/sEstimated Auto-generated benchmark | 4GB |
| Qwen/Qwen3-14B | Q8 | 329.15 tok/sEstimated Auto-generated benchmark | 14GB |
| ibm-research/PowerMoE-3b | FP16 | 328.04 tok/sEstimated Auto-generated benchmark | 6GB |
| meta-llama/Llama-3.2-1B | FP16 | 327.67 tok/sEstimated Auto-generated benchmark | 2GB |
| google/gemma-3-1b-it | FP16 | 321.89 tok/sEstimated Auto-generated benchmark | 2GB |
| meta-llama/Llama-3.2-1B-Instruct | FP16 | 320.33 tok/sEstimated Auto-generated benchmark | 2GB |
| unsloth/gemma-3-1b-it | FP16 | 317.47 tok/sEstimated Auto-generated benchmark | 2GB |
| bigcode/starcoder2-3b | FP16 | 316.41 tok/sEstimated Auto-generated benchmark | 6GB |
| context-labs/meta-llama-Llama-3.2-3B-Instruct-FP16 | FP16 | 314.98 tok/sEstimated Auto-generated benchmark | 6GB |
| nari-labs/Dia2-2B | FP16 | 312.74 tok/sEstimated Auto-generated benchmark | 5GB |
| deepseek-ai/DeepSeek-OCR | FP16 | 309.13 tok/sEstimated Auto-generated benchmark | 7GB |
| google/gemma-2b | FP16 | 307.94 tok/sEstimated Auto-generated benchmark | 4GB |
| tencent/HunyuanOCR | FP16 | 305.41 tok/sEstimated Auto-generated benchmark | 3GB |
| Qwen/Qwen2.5-3B | FP16 | 303.95 tok/sEstimated Auto-generated benchmark | 6GB |
| google/gemma-2-2b-it | FP16 | 301.84 tok/sEstimated Auto-generated benchmark | 4GB |
| inference-net/Schematron-3B | FP16 | 301.30 tok/sEstimated Auto-generated benchmark | 6GB |
| google/embeddinggemma-300m | FP16 | 300.77 tok/sEstimated Auto-generated benchmark | 1GB |
| Qwen/Qwen2.5-3B-Instruct | FP16 | 298.95 tok/sEstimated Auto-generated benchmark | 6GB |
| google-bert/bert-base-uncased | FP16 | 296.01 tok/sEstimated Auto-generated benchmark | 1GB |
| meta-llama/Llama-3.2-3B | FP16 | 293.43 tok/sEstimated Auto-generated benchmark | 6GB |
| meta-llama/Llama-3.2-3B-Instruct | FP16 | 293.03 tok/sEstimated Auto-generated benchmark | 6GB |
| TinyLlama/TinyLlama-1.1B-Chat-v1.0 | FP16 | 290.22 tok/sEstimated Auto-generated benchmark | 2GB |
| facebook/sam3 | FP16 | 287.77 tok/sEstimated Auto-generated benchmark | 2GB |
| openai-community/gpt2-large | FP16 | 287.45 tok/sEstimated Auto-generated benchmark | 15GB |
| Qwen/Qwen2.5-Coder-1.5B | FP16 | 287.15 tok/sEstimated Auto-generated benchmark | 11GB |
| lmstudio-community/DeepSeek-R1-0528-Qwen3-8B-MLX-8bit | FP16 | 287.05 tok/sEstimated Auto-generated benchmark | 17GB |
| meta-llama/Llama-Guard-3-1B | FP16 | 286.77 tok/sEstimated Auto-generated benchmark | 2GB |
| unsloth/Llama-3.2-1B-Instruct | FP16 | 286.70 tok/sEstimated Auto-generated benchmark | 2GB |
| deepseek-ai/DeepSeek-V3-0324 | FP16 | 286.42 tok/sEstimated Auto-generated benchmark | 15GB |
| Qwen/Qwen3-0.6B-Base | FP16 | 286.30 tok/sEstimated Auto-generated benchmark | 13GB |
| microsoft/Phi-3-mini-4k-instruct | FP16 | 286.21 tok/sEstimated Auto-generated benchmark | 15GB |
| microsoft/Phi-3-mini-128k-instruct | FP16 | 286.11 tok/sEstimated Auto-generated benchmark | 15GB |
| Qwen/Qwen3-4B-Instruct-2507 | FP16 | 285.94 tok/sEstimated Auto-generated benchmark | 9GB |
| EleutherAI/gpt-neo-125m | FP16 | 284.68 tok/sEstimated Auto-generated benchmark | 15GB |
| microsoft/DialoGPT-small | FP16 | 284.51 tok/sEstimated Auto-generated benchmark | 15GB |
| meta-llama/Meta-Llama-3-8B-Instruct | FP16 | 284.48 tok/sEstimated Auto-generated benchmark | 17GB |
| Qwen/Qwen2-0.5B | FP16 | 284.45 tok/sEstimated Auto-generated benchmark | 11GB |
| kaitchup/Phi-3-mini-4k-instruct-gptq-4bit | FP16 | 284.37 tok/sEstimated Auto-generated benchmark | 9GB |
| Qwen/Qwen3-30B-A3B-Instruct-2507 | Q8 | 283.83 tok/sEstimated Auto-generated benchmark | 31GB |
| Qwen/Qwen2.5-7B | FP16 | 283.82 tok/sEstimated Auto-generated benchmark | 15GB |
| lmstudio-community/Qwen3-Coder-30B-A3B-Instruct-MLX-8bit | Q8 | 283.81 tok/sEstimated Auto-generated benchmark | 31GB |
| rednote-hilab/dots.ocr | FP16 | 283.60 tok/sEstimated Auto-generated benchmark | 15GB |
| Qwen/Qwen2.5-Coder-7B-Instruct | FP16 | 283.46 tok/sEstimated Auto-generated benchmark | 15GB |
| google/gemma-3-270m-it | FP16 | 283.44 tok/sEstimated Auto-generated benchmark | 15GB |
| Qwen/Qwen3-Reranker-0.6B | FP16 | 283.01 tok/sEstimated Auto-generated benchmark | 13GB |
| Qwen/Qwen3-1.7B-Base | FP16 | 282.69 tok/sEstimated Auto-generated benchmark | 15GB |
| deepseek-ai/DeepSeek-V3.1 | FP16 | 282.24 tok/sEstimated Auto-generated benchmark | 15GB |
| openai/gpt-oss-20b | Q8 | 282.21 tok/sEstimated Auto-generated benchmark | 20GB |
| trl-internal-testing/tiny-random-LlamaForCausalLM | FP16 | 281.89 tok/sEstimated Auto-generated benchmark | 15GB |
| Qwen/Qwen3-30B-A3B | Q8 | 281.25 tok/sEstimated Auto-generated benchmark | 31GB |
| Qwen/Qwen3-Coder-30B-A3B-Instruct | Q8 | 280.85 tok/sEstimated Auto-generated benchmark | 31GB |
| unsloth/gpt-oss-20b-unsloth-bnb-4bit | Q8 | 280.42 tok/sEstimated Auto-generated benchmark | 20GB |
| meta-llama/Llama-Guard-3-8B | FP16 | 280.28 tok/sEstimated Auto-generated benchmark | 17GB |
| microsoft/Phi-3.5-mini-instruct | FP16 | 280.28 tok/sEstimated Auto-generated benchmark | 15GB |
| Tongyi-MAI/Z-Image-Turbo | FP16 | 280.22 tok/sEstimated Auto-generated benchmark | 16GB |
| tencent/HunyuanVideo-1.5 | FP16 | 279.06 tok/sEstimated Auto-generated benchmark | 16GB |
| bigscience/bloomz-560m | FP16 | 278.98 tok/sEstimated Auto-generated benchmark | 15GB |
| sshleifer/tiny-gpt2 | FP16 | 278.32 tok/sEstimated Auto-generated benchmark | 15GB |
| Qwen/Qwen-Image-Edit-2509 | FP16 | 277.92 tok/sEstimated Auto-generated benchmark | 16GB |
| deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct | FP16 | 277.78 tok/sEstimated Auto-generated benchmark | 15GB |
| rinna/japanese-gpt-neox-small | FP16 | 277.44 tok/sEstimated Auto-generated benchmark | 15GB |
| microsoft/VibeVoice-1.5B | FP16 | 276.95 tok/sEstimated Auto-generated benchmark | 11GB |
| unsloth/mistral-7b-v0.3-bnb-4bit | FP16 | 276.65 tok/sEstimated Auto-generated benchmark | 15GB |
| unsloth/gpt-oss-20b-BF16 | Q8 | 275.82 tok/sEstimated Auto-generated benchmark | 20GB |
| trl-internal-testing/tiny-Qwen2ForCausalLM-2.5 | FP16 | 275.74 tok/sEstimated Auto-generated benchmark | 15GB |
| meta-llama/Meta-Llama-3-8B | FP16 | 274.18 tok/sEstimated Auto-generated benchmark | 17GB |
| lmstudio-community/Qwen3-Coder-30B-A3B-Instruct-MLX-5bit | Q8 | 273.92 tok/sEstimated Auto-generated benchmark | 31GB |
| numind/NuExtract-1.5 | FP16 | 273.57 tok/sEstimated Auto-generated benchmark | 15GB |
| microsoft/Phi-3.5-mini-instruct | FP16 | 273.42 tok/sEstimated Auto-generated benchmark | 8GB |
| deepseek-ai/DeepSeek-R1 | FP16 | 273.41 tok/sEstimated Auto-generated benchmark | 15GB |
| Qwen/Qwen2.5-Math-1.5B | FP16 | 273.11 tok/sEstimated Auto-generated benchmark | 11GB |
| Qwen/Qwen3-Embedding-8B | FP16 | 272.85 tok/sEstimated Auto-generated benchmark | 17GB |
| meta-llama/Llama-3.1-8B-Instruct | FP16 | 271.22 tok/sEstimated Auto-generated benchmark | 17GB |
| BSC-LT/salamandraTA-7b-instruct | FP16 | 270.60 tok/sEstimated Auto-generated benchmark | 15GB |
| facebook/opt-125m | FP16 | 270.54 tok/sEstimated Auto-generated benchmark | 15GB |
| Gensyn/Qwen2.5-0.5B-Instruct | FP16 | 270.43 tok/sEstimated Auto-generated benchmark | 11GB |
| Qwen/Qwen2.5-0.5B | FP16 | 269.83 tok/sEstimated Auto-generated benchmark | 11GB |
| swiss-ai/Apertus-8B-Instruct-2509 | FP16 | 269.64 tok/sEstimated Auto-generated benchmark | 17GB |
| deepseek-ai/DeepSeek-R1-Distill-Qwen-7B | FP16 | 269.26 tok/sEstimated Auto-generated benchmark | 15GB |
| meta-llama/Llama-2-7b-chat-hf | FP16 | 268.96 tok/sEstimated Auto-generated benchmark | 15GB |
| HuggingFaceH4/zephyr-7b-beta | FP16 | 268.78 tok/sEstimated Auto-generated benchmark | 15GB |
| GSAI-ML/LLaDA-8B-Instruct | FP16 | 268.41 tok/sEstimated Auto-generated benchmark | 17GB |
| GSAI-ML/LLaDA-8B-Base | FP16 | 267.69 tok/sEstimated Auto-generated benchmark | 17GB |
| black-forest-labs/FLUX.2-dev | FP16 | 267.55 tok/sEstimated Auto-generated benchmark | 16GB |
| zai-org/GLM-4.6-FP8 | FP16 | 266.21 tok/sEstimated Auto-generated benchmark | 15GB |
| llamafactory/tiny-random-Llama-3 | FP16 | 265.90 tok/sEstimated Auto-generated benchmark | 15GB |
| skt/kogpt2-base-v2 | FP16 | 265.45 tok/sEstimated Auto-generated benchmark | 15GB |
| black-forest-labs/FLUX.1-dev | FP16 | 265.19 tok/sEstimated Auto-generated benchmark | 16GB |
| distilbert/distilgpt2 | FP16 | 264.44 tok/sEstimated Auto-generated benchmark | 15GB |
| deepseek-ai/DeepSeek-V2.5 | Q4 | 264.44 tok/sEstimated Auto-generated benchmark | 328GB |
| lmstudio-community/Qwen3-Coder-30B-A3B-Instruct-MLX-4bit | Q8 | 264.44 tok/sEstimated Auto-generated benchmark | 31GB |
| ibm-granite/granite-3.3-8b-instruct | FP16 | 263.58 tok/sEstimated Auto-generated benchmark | 17GB |
| Qwen/Qwen3-4B-Thinking-2507-FP8 | FP16 | 263.44 tok/sEstimated Auto-generated benchmark | 9GB |
| Qwen/Qwen3-Embedding-0.6B | FP16 | 263.14 tok/sEstimated Auto-generated benchmark | 13GB |
| HuggingFaceTB/SmolLM2-135M | FP16 | 262.30 tok/sEstimated Auto-generated benchmark | 15GB |
| lmstudio-community/DeepSeek-R1-0528-Qwen3-8B-MLX-4bit | FP16 | 262.15 tok/sEstimated Auto-generated benchmark | 17GB |
| huggyllama/llama-7b | FP16 | 262.07 tok/sEstimated Auto-generated benchmark | 15GB |
| lmstudio-community/Qwen3-Coder-30B-A3B-Instruct-MLX-6bit | Q8 | 261.80 tok/sEstimated Auto-generated benchmark | 31GB |
| microsoft/Phi-4-multimodal-instruct | FP16 | 261.41 tok/sEstimated Auto-generated benchmark | 15GB |
| Qwen/Qwen2.5-7B-Instruct | FP16 | 261.30 tok/sEstimated Auto-generated benchmark | 16GB |
| hmellor/tiny-random-LlamaForCausalLM | FP16 | 260.85 tok/sEstimated Auto-generated benchmark | 15GB |
| lmsys/vicuna-7b-v1.5 | FP16 | 260.56 tok/sEstimated Auto-generated benchmark | 15GB |
| lmstudio-community/Qwen3-4B-Thinking-2507-MLX-8bit | FP16 | 260.42 tok/sEstimated Auto-generated benchmark | 9GB |
| Qwen/Qwen2-7B-Instruct | FP16 | 260.35 tok/sEstimated Auto-generated benchmark | 15GB |
| moonshotai/Kimi-K2-Thinking | Q4 | 259.92 tok/sEstimated Auto-generated benchmark | 489GB |
| Qwen/Qwen3-0.6B | FP16 | 259.40 tok/sEstimated Auto-generated benchmark | 13GB |
| deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B | FP16 | 258.50 tok/sEstimated Auto-generated benchmark | 11GB |
| openai/gpt-oss-safeguard-20b | Q8 | 257.60 tok/sEstimated Auto-generated benchmark | 22GB |
| Qwen/Qwen2.5-32B-Instruct | Q4 | 257.35 tok/sEstimated Auto-generated benchmark | 17GB |
| IlyaGusev/saiga_llama3_8b | FP16 | 257.33 tok/sEstimated Auto-generated benchmark | 17GB |
| trl-internal-testing/tiny-LlamaForCausalLM-3.2 | FP16 | 257.16 tok/sEstimated Auto-generated benchmark | 15GB |
| mistralai/Mistral-Small-Instruct-2409 | Q8 | 256.93 tok/sEstimated Auto-generated benchmark | 23GB |
| baichuan-inc/Baichuan-M2-32B | Q4 | 256.80 tok/sEstimated Auto-generated benchmark | 16GB |
| Qwen/Qwen3-8B-FP8 | FP16 | 256.70 tok/sEstimated Auto-generated benchmark | 17GB |
| openai-community/gpt2-medium | FP16 | 256.52 tok/sEstimated Auto-generated benchmark | 15GB |
| Alibaba-NLP/gte-Qwen2-1.5B-instruct | FP16 | 256.52 tok/sEstimated Auto-generated benchmark | 11GB |
| deepseek-ai/deepseek-coder-33b-instruct | Q4 | 256.47 tok/sEstimated Auto-generated benchmark | 17GB |
| mistralai/Mistral-7B-Instruct-v0.2 | FP16 | 256.25 tok/sEstimated Auto-generated benchmark | 15GB |
| HuggingFaceM4/tiny-random-LlamaForCausalLM | FP16 | 256.00 tok/sEstimated Auto-generated benchmark | 15GB |
| Qwen/Qwen2-0.5B-Instruct | FP16 | 254.89 tok/sEstimated Auto-generated benchmark | 11GB |
| Qwen/Qwen3-Embedding-4B | FP16 | 254.40 tok/sEstimated Auto-generated benchmark | 9GB |
| petals-team/StableBeluga2 | FP16 | 254.34 tok/sEstimated Auto-generated benchmark | 15GB |
| Qwen/Qwen2.5-0.5B-Instruct | FP16 | 254.07 tok/sEstimated Auto-generated benchmark | 11GB |
| Qwen/Qwen2.5-7B-Instruct | FP16 | 253.57 tok/sEstimated Auto-generated benchmark | 15GB |
| allenai/Olmo-3-7B-Think | FP16 | 253.41 tok/sEstimated Auto-generated benchmark | 16GB |
| Qwen/Qwen2-1.5B-Instruct | FP16 | 253.37 tok/sEstimated Auto-generated benchmark | 11GB |
| mlx-community/gpt-oss-20b-MXFP4-Q8 | Q8 | 253.32 tok/sEstimated Auto-generated benchmark | 20GB |
| ibm-granite/granite-docling-258M | FP16 | 253.01 tok/sEstimated Auto-generated benchmark | 15GB |
| deepseek-ai/DeepSeek-V3 | FP16 | 252.93 tok/sEstimated Auto-generated benchmark | 15GB |
| RedHatAI/Llama-3.3-70B-Instruct-FP8-dynamic | Q4 | 252.89 tok/sEstimated Auto-generated benchmark | 34GB |
| microsoft/phi-2 | FP16 | 252.73 tok/sEstimated Auto-generated benchmark | 15GB |
| Qwen/Qwen3-4B-Base | FP16 | 252.55 tok/sEstimated Auto-generated benchmark | 9GB |
| deepseek-ai/DeepSeek-R1-Distill-Llama-8B | FP16 | 252.45 tok/sEstimated Auto-generated benchmark | 17GB |
| lmstudio-community/Qwen3-4B-Thinking-2507-MLX-4bit | FP16 | 252.32 tok/sEstimated Auto-generated benchmark | 9GB |
| Qwen/Qwen3-8B | FP16 | 252.19 tok/sEstimated Auto-generated benchmark | 17GB |
| meta-llama/Llama-2-7b-hf | FP16 | 252.11 tok/sEstimated Auto-generated benchmark | 15GB |
| deepseek-ai/DeepSeek-R1-0528 | FP16 | 251.83 tok/sEstimated Auto-generated benchmark | 15GB |
| Qwen/Qwen3-30B-A3B-Thinking-2507 | Q8 | 251.52 tok/sEstimated Auto-generated benchmark | 31GB |
| Qwen/Qwen2.5-32B | Q4 | 251.00 tok/sEstimated Auto-generated benchmark | 16GB |
| NousResearch/Meta-Llama-3.1-8B-Instruct | FP16 | 250.50 tok/sEstimated Auto-generated benchmark | 17GB |
| Qwen/Qwen2.5-1.5B | FP16 | 250.45 tok/sEstimated Auto-generated benchmark | 11GB |
| zai-org/GLM-4.5-Air | FP16 | 249.97 tok/sEstimated Auto-generated benchmark | 15GB |
| google/gemma-2-27b-it | Q8 | 249.16 tok/sEstimated Auto-generated benchmark | 28GB |
| microsoft/DialoGPT-medium | FP16 | 248.85 tok/sEstimated Auto-generated benchmark | 15GB |
| meta-llama/Llama-3.3-70B-Instruct | Q4 | 248.64 tok/sEstimated Auto-generated benchmark | 34GB |
| microsoft/phi-4 | FP16 | 248.26 tok/sEstimated Auto-generated benchmark | 15GB |
| moonshotai/Kimi-Linear-48B-A3B-Instruct | Q4 | 248.00 tok/sEstimated Auto-generated benchmark | 25GB |
| microsoft/Phi-4-mini-instruct | FP16 | 246.84 tok/sEstimated Auto-generated benchmark | 15GB |
| MiniMaxAI/MiniMax-M2 | FP16 | 246.77 tok/sEstimated Auto-generated benchmark | 15GB |
| 01-ai/Yi-1.5-34B-Chat | Q4 | 246.29 tok/sEstimated Auto-generated benchmark | 18GB |
| unsloth/Meta-Llama-3.1-8B-Instruct | FP16 | 246.02 tok/sEstimated Auto-generated benchmark | 17GB |
| dphn/dolphin-2.9.1-yi-1.5-34b | Q4 | 245.52 tok/sEstimated Auto-generated benchmark | 17GB |
| unsloth/Meta-Llama-3.1-8B-Instruct-bnb-4bit | FP16 | 245.43 tok/sEstimated Auto-generated benchmark | 17GB |
| Qwen/Qwen3-8B-Base | FP16 | 244.24 tok/sEstimated Auto-generated benchmark | 17GB |
| microsoft/Phi-3.5-vision-instruct | FP16 | 244.09 tok/sEstimated Auto-generated benchmark | 15GB |
| Qwen/Qwen3-4B | FP16 | 243.42 tok/sEstimated Auto-generated benchmark | 9GB |
| openai-community/gpt2 | FP16 | 243.32 tok/sEstimated Auto-generated benchmark | 15GB |
| openai-community/gpt2-xl | FP16 | 243.13 tok/sEstimated Auto-generated benchmark | 15GB |
| meta-llama/Llama-3.2-3B-Instruct | FP16 | 243.00 tok/sEstimated Auto-generated benchmark | 7GB |
| Qwen/QwQ-32B-Preview | Q4 | 242.82 tok/sEstimated Auto-generated benchmark | 17GB |
| Qwen/Qwen3-30B-A3B-Instruct-2507-FP8 | Q8 | 242.31 tok/sEstimated Auto-generated benchmark | 31GB |
| parler-tts/parler-tts-large-v1 | FP16 | 242.03 tok/sEstimated Auto-generated benchmark | 15GB |
| Qwen/Qwen3-4B-Thinking-2507 | FP16 | 239.64 tok/sEstimated Auto-generated benchmark | 9GB |
| meta-llama/Meta-Llama-3-70B-Instruct | Q4 | 239.28 tok/sEstimated Auto-generated benchmark | 34GB |
| deepseek-ai/DeepSeek-R1-Distill-Qwen-32B | Q4 | 238.80 tok/sEstimated Auto-generated benchmark | 16GB |
| EleutherAI/pythia-70m-deduped | FP16 | 238.77 tok/sEstimated Auto-generated benchmark | 15GB |
| liuhaotian/llava-v1.5-7b | FP16 | 238.25 tok/sEstimated Auto-generated benchmark | 15GB |
| Qwen/Qwen3-1.7B | FP16 | 238.22 tok/sEstimated Auto-generated benchmark | 15GB |
| mistralai/Mistral-7B-v0.1 | FP16 | 237.55 tok/sEstimated Auto-generated benchmark | 15GB |
| dicta-il/dictalm2.0-instruct | FP16 | 237.11 tok/sEstimated Auto-generated benchmark | 15GB |
| Qwen/Qwen2.5-1.5B-Instruct | FP16 | 236.78 tok/sEstimated Auto-generated benchmark | 11GB |
| vikhyatk/moondream2 | FP16 | 236.36 tok/sEstimated Auto-generated benchmark | 15GB |
| lmstudio-community/Qwen3-4B-Thinking-2507-MLX-6bit | FP16 | 236.30 tok/sEstimated Auto-generated benchmark | 9GB |
| HuggingFaceTB/SmolLM-135M | FP16 | 236.18 tok/sEstimated Auto-generated benchmark | 15GB |
| meta-llama/Llama-3.1-8B | FP16 | 236.17 tok/sEstimated Auto-generated benchmark | 17GB |
| mistralai/Mistral-7B-Instruct-v0.1 | FP16 | 235.97 tok/sEstimated Auto-generated benchmark | 15GB |
| Qwen/Qwen2.5-32B-Instruct | Q4 | 231.75 tok/sEstimated Auto-generated benchmark | 16GB |
| RedHatAI/Meta-Llama-3.1-70B-Instruct-quantized.w4a16 | Q4 | 224.85 tok/sEstimated Auto-generated benchmark | 34GB |
| Qwen/Qwen3-32B | Q4 | 221.67 tok/sEstimated Auto-generated benchmark | 16GB |
| Qwen/Qwen2.5-Coder-32B-Instruct | Q4 | 221.57 tok/sEstimated Auto-generated benchmark | 17GB |
| unsloth/DeepSeek-R1-Distill-Qwen-32B-bnb-4bit | Q4 | 220.00 tok/sEstimated Auto-generated benchmark | 16GB |
| codellama/CodeLlama-34b-hf | Q4 | 217.90 tok/sEstimated Auto-generated benchmark | 17GB |
| meta-llama/Llama-3.1-70B-Instruct | Q4 | 217.15 tok/sEstimated Auto-generated benchmark | 34GB |
| Qwen/Qwen2.5-14B-Instruct | FP16 | 216.08 tok/sEstimated Auto-generated benchmark | 29GB |
| Qwen/Qwen2.5-14B | FP16 | 213.10 tok/sEstimated Auto-generated benchmark | 29GB |
| nvidia/NVIDIA-Nemotron-Nano-9B-v2 | FP16 | 209.04 tok/sEstimated Auto-generated benchmark | 19GB |
| microsoft/Phi-3-medium-128k-instruct | FP16 | 200.91 tok/sEstimated Auto-generated benchmark | 29GB |
| NousResearch/Hermes-3-Llama-3.1-8B | FP16 | 195.62 tok/sEstimated Auto-generated benchmark | 17GB |
| meta-llama/Llama-2-13b-chat-hf | FP16 | 195.47 tok/sEstimated Auto-generated benchmark | 27GB |
| ai-forever/ruGPT-3.5-13B | FP16 | 195.13 tok/sEstimated Auto-generated benchmark | 27GB |
| Qwen/Qwen3-14B-Base | FP16 | 193.89 tok/sEstimated Auto-generated benchmark | 29GB |
| EssentialAI/rnj-1 | FP16 | 192.38 tok/sEstimated Auto-generated benchmark | 19GB |
| meta-llama/Llama-3.1-8B-Instruct | FP16 | 186.97 tok/sEstimated Auto-generated benchmark | 17GB |
| mistralai/Ministral-3-14B-Instruct-2512 | FP16 | 185.49 tok/sEstimated Auto-generated benchmark | 32GB |
| meta-llama/Llama-3.3-70B-Instruct | Q8 | 182.50 tok/sEstimated Auto-generated benchmark | 68GB |
| Qwen/Qwen3-14B | FP16 | 182.03 tok/sEstimated Auto-generated benchmark | 29GB |
| deepseek-ai/DeepSeek-V2.5 | Q8 | 181.09 tok/sEstimated Auto-generated benchmark | 656GB |
| OpenPipe/Qwen3-14B-Instruct | FP16 | 181.04 tok/sEstimated Auto-generated benchmark | 29GB |
| dphn/dolphin-2.9.1-yi-1.5-34b | Q8 | 180.63 tok/sEstimated Auto-generated benchmark | 35GB |
| moonshotai/Kimi-K2-Thinking | Q8 | 180.17 tok/sEstimated Auto-generated benchmark | 978GB |
| Qwen/Qwen2.5-14B-Instruct | FP16 | 177.81 tok/sEstimated Auto-generated benchmark | 30GB |
| Qwen/Qwen2.5-Coder-32B-Instruct | Q8 | 177.66 tok/sEstimated Auto-generated benchmark | 34GB |
| google/gemma-2-9b-it | FP16 | 176.85 tok/sEstimated Auto-generated benchmark | 20GB |
| meta-llama/Llama-3.1-70B-Instruct | Q8 | 176.03 tok/sEstimated Auto-generated benchmark | 68GB |
| Qwen/Qwen2.5-32B | Q8 | 174.87 tok/sEstimated Auto-generated benchmark | 33GB |
| Qwen/Qwen2.5-32B-Instruct | Q8 | 174.17 tok/sEstimated Auto-generated benchmark | 34GB |
| RedHatAI/Llama-3.3-70B-Instruct-FP8-dynamic | Q8 | 174.12 tok/sEstimated Auto-generated benchmark | 68GB |
| Qwen/Qwen3-32B | Q8 | 172.60 tok/sEstimated Auto-generated benchmark | 33GB |
| Qwen/QwQ-32B-Preview | Q8 | 170.89 tok/sEstimated Auto-generated benchmark | 34GB |
| mistralai/Mixtral-8x22B-Instruct-v0.1 | Q4 | 166.80 tok/sEstimated Auto-generated benchmark | 69GB |
| Qwen/Qwen2.5-32B-Instruct | Q8 | 166.66 tok/sEstimated Auto-generated benchmark | 33GB |
| 01-ai/Yi-1.5-34B-Chat | Q8 | 164.11 tok/sEstimated Auto-generated benchmark | 35GB |
| baichuan-inc/Baichuan-M2-32B | Q8 | 163.75 tok/sEstimated Auto-generated benchmark | 33GB |
| meta-llama/Meta-Llama-3-70B-Instruct | Q8 | 161.09 tok/sEstimated Auto-generated benchmark | 68GB |
| unsloth/DeepSeek-R1-Distill-Qwen-32B-bnb-4bit | Q8 | 159.83 tok/sEstimated Auto-generated benchmark | 33GB |
| deepseek-ai/deepseek-coder-33b-instruct | Q8 | 158.60 tok/sEstimated Auto-generated benchmark | 34GB |
| RedHatAI/Meta-Llama-3.1-70B-Instruct-quantized.w4a16 | Q8 | 157.39 tok/sEstimated Auto-generated benchmark | 68GB |
| Qwen/Qwen3-30B-A3B-Thinking-2507 | FP16 | 157.11 tok/sEstimated Auto-generated benchmark | 61GB |
| deepseek-ai/DeepSeek-R1-Distill-Qwen-32B | Q8 | 156.61 tok/sEstimated Auto-generated benchmark | 33GB |
| moonshotai/Kimi-Linear-48B-A3B-Instruct | Q8 | 154.28 tok/sEstimated Auto-generated benchmark | 50GB |
| openai/gpt-oss-20b | FP16 | 154.22 tok/sEstimated Auto-generated benchmark | 41GB |
| unsloth/gpt-oss-20b-unsloth-bnb-4bit | FP16 | 153.36 tok/sEstimated Auto-generated benchmark | 41GB |
| codellama/CodeLlama-34b-hf | Q8 | 152.19 tok/sEstimated Auto-generated benchmark | 35GB |
| Qwen/Qwen3-30B-A3B-Instruct-2507-FP8 | FP16 | 151.65 tok/sEstimated Auto-generated benchmark | 61GB |
| Qwen/Qwen2.5-Math-72B-Instruct | Q4 | 151.40 tok/sEstimated Auto-generated benchmark | 36GB |
| openai/gpt-oss-safeguard-20b | FP16 | 150.62 tok/sEstimated Auto-generated benchmark | 44GB |
| meta-llama/Llama-3.1-70B-Instruct | Q4 | 149.17 tok/sEstimated Auto-generated benchmark | 34GB |
| Qwen/Qwen3-Coder-30B-A3B-Instruct | FP16 | 147.45 tok/sEstimated Auto-generated benchmark | 61GB |
| lmstudio-community/Qwen3-Coder-30B-A3B-Instruct-MLX-5bit | FP16 | 144.43 tok/sEstimated Auto-generated benchmark | 61GB |
| Qwen/Qwen3-30B-A3B-Instruct-2507 | FP16 | 144.42 tok/sEstimated Auto-generated benchmark | 61GB |
| lmstudio-community/Qwen3-Coder-30B-A3B-Instruct-MLX-8bit | FP16 | 144.09 tok/sEstimated Auto-generated benchmark | 61GB |
| nvidia/Llama-3.1-Nemotron-70B-Instruct-HF | Q4 | 144.09 tok/sEstimated Auto-generated benchmark | 34GB |
| Qwen/Qwen3-Next-80B-A3B-Instruct-FP8 | Q4 | 142.40 tok/sEstimated Auto-generated benchmark | 39GB |
| unsloth/gpt-oss-20b-BF16 | FP16 | 141.42 tok/sEstimated Auto-generated benchmark | 41GB |
| Qwen/Qwen2.5-72B-Instruct | Q4 | 141.11 tok/sEstimated Auto-generated benchmark | 35GB |
| lmstudio-community/Qwen3-Coder-30B-A3B-Instruct-MLX-6bit | FP16 | 140.56 tok/sEstimated Auto-generated benchmark | 61GB |
| openai/gpt-oss-120b | Q4 | 140.16 tok/sEstimated Auto-generated benchmark | 59GB |
| google/gemma-2-27b-it | FP16 | 139.57 tok/sEstimated Auto-generated benchmark | 56GB |
| NousResearch/Hermes-3-Llama-3.1-70B | Q4 | 138.60 tok/sEstimated Auto-generated benchmark | 34GB |
| mistralai/Mistral-Large-Instruct-2411 | Q4 | 138.45 tok/sEstimated Auto-generated benchmark | 60GB |
| mistralai/Mixtral-8x22B-Instruct-v0.1 | Q8 | 137.87 tok/sEstimated Auto-generated benchmark | 138GB |
| mlx-community/gpt-oss-20b-MXFP4-Q8 | FP16 | 137.40 tok/sEstimated Auto-generated benchmark | 41GB |
| lmstudio-community/Qwen3-Coder-30B-A3B-Instruct-MLX-4bit | FP16 | 135.57 tok/sEstimated Auto-generated benchmark | 61GB |
| mistralai/Mistral-Small-Instruct-2409 | FP16 | 134.44 tok/sEstimated Auto-generated benchmark | 46GB |
| Qwen/Qwen3-Next-80B-A3B-Thinking | Q4 | 134.40 tok/sEstimated Auto-generated benchmark | 39GB |
| Qwen/Qwen3-Next-80B-A3B-Thinking-FP8 | Q4 | 133.83 tok/sEstimated Auto-generated benchmark | 39GB |
| Qwen/Qwen3-Next-80B-A3B-Instruct | Q4 | 131.67 tok/sEstimated Auto-generated benchmark | 39GB |
| Qwen/Qwen3-30B-A3B | FP16 | 131.16 tok/sEstimated Auto-generated benchmark | 61GB |
| AI-MO/Kimina-Prover-72B | Q4 | 128.03 tok/sEstimated Auto-generated benchmark | 35GB |
| meta-llama/Llama-3.3-70B-Instruct | Q4 | 127.63 tok/sEstimated Auto-generated benchmark | 34GB |
| Qwen/Qwen2.5-72B-Instruct | Q4 | 126.95 tok/sEstimated Auto-generated benchmark | 36GB |
| RedHatAI/Llama-3.2-90B-Vision-Instruct-FP8-dynamic | Q4 | 124.45 tok/sEstimated Auto-generated benchmark | 44GB |
| deepseek-ai/DeepSeek-Coder-V2-Instruct-0724 | Q4 | 110.38 tok/sEstimated Auto-generated benchmark | 115GB |
| Qwen/Qwen2.5-72B-Instruct | Q8 | 105.82 tok/sEstimated Auto-generated benchmark | 70GB |
| Qwen/Qwen3-Next-80B-A3B-Thinking | Q8 | 105.74 tok/sEstimated Auto-generated benchmark | 78GB |
| openai/gpt-oss-120b | Q8 | 104.25 tok/sEstimated Auto-generated benchmark | 117GB |
| Qwen/Qwen3-Next-80B-A3B-Instruct | Q8 | 103.73 tok/sEstimated Auto-generated benchmark | 78GB |
| AI-MO/Kimina-Prover-72B | Q8 | 102.33 tok/sEstimated Auto-generated benchmark | 70GB |
| meta-llama/Llama-3.3-70B-Instruct | Q8 | 101.72 tok/sEstimated Auto-generated benchmark | 69GB |
| RedHatAI/Llama-3.2-90B-Vision-Instruct-FP8-dynamic | Q8 | 100.96 tok/sEstimated Auto-generated benchmark | 88GB |
| mistralai/Mistral-Large-Instruct-2411 | Q8 | 100.32 tok/sEstimated Auto-generated benchmark | 120GB |
| deepseek-ai/DeepSeek-Math-V2 | Q4 | 99.92 tok/sEstimated Auto-generated benchmark | 383GB |
| meta-llama/Llama-3.1-70B-Instruct | FP16 | 98.73 tok/sEstimated Auto-generated benchmark | 137GB |
| Qwen/Qwen2.5-Coder-32B-Instruct | FP16 | 98.64 tok/sEstimated Auto-generated benchmark | 67GB |
| unsloth/DeepSeek-R1-Distill-Qwen-32B-bnb-4bit | FP16 | 98.49 tok/sEstimated Auto-generated benchmark | 66GB |
| 01-ai/Yi-1.5-34B-Chat | FP16 | 98.29 tok/sEstimated Auto-generated benchmark | 70GB |
| Qwen/QwQ-32B-Preview | FP16 | 97.91 tok/sEstimated Auto-generated benchmark | 67GB |
| meta-llama/Meta-Llama-3-70B-Instruct | FP16 | 97.61 tok/sEstimated Auto-generated benchmark | 137GB |
| nvidia/Llama-3.1-Nemotron-70B-Instruct-HF | Q8 | 96.56 tok/sEstimated Auto-generated benchmark | 69GB |
| meta-llama/Llama-3.1-70B-Instruct | Q8 | 95.97 tok/sEstimated Auto-generated benchmark | 69GB |
| Qwen/Qwen2.5-32B-Instruct | FP16 | 95.53 tok/sEstimated Auto-generated benchmark | 67GB |
| Qwen/Qwen2.5-72B-Instruct | Q8 | 94.93 tok/sEstimated Auto-generated benchmark | 71GB |
| Qwen/Qwen3-Next-80B-A3B-Thinking-FP8 | Q8 | 94.36 tok/sEstimated Auto-generated benchmark | 78GB |
| Qwen/Qwen2.5-Math-72B-Instruct | Q8 | 94.34 tok/sEstimated Auto-generated benchmark | 71GB |
| RedHatAI/Llama-3.3-70B-Instruct-FP8-dynamic | FP16 | 94.22 tok/sEstimated Auto-generated benchmark | 137GB |
| dphn/dolphin-2.9.1-yi-1.5-34b | FP16 | 93.92 tok/sEstimated Auto-generated benchmark | 70GB |
| deepseek-ai/DeepSeek-R1-Distill-Qwen-32B | FP16 | 92.49 tok/sEstimated Auto-generated benchmark | 66GB |
| RedHatAI/Meta-Llama-3.1-70B-Instruct-quantized.w4a16 | FP16 | 91.67 tok/sEstimated Auto-generated benchmark | 137GB |
| Qwen/Qwen3-Next-80B-A3B-Instruct-FP8 | Q8 | 91.54 tok/sEstimated Auto-generated benchmark | 78GB |
| baichuan-inc/Baichuan-M2-32B | FP16 | 91.43 tok/sEstimated Auto-generated benchmark | 66GB |
| Qwen/Qwen2.5-32B | FP16 | 90.36 tok/sEstimated Auto-generated benchmark | 66GB |
| deepseek-ai/DeepSeek-V2.5 | FP16 | 89.13 tok/sEstimated Auto-generated benchmark | 1312GB |
| NousResearch/Hermes-3-Llama-3.1-70B | Q8 | 88.02 tok/sEstimated Auto-generated benchmark | 69GB |
| Qwen/Qwen3-235B-A22B | Q4 | 87.23 tok/sEstimated Auto-generated benchmark | 115GB |
| codellama/CodeLlama-34b-hf | FP16 | 85.74 tok/sEstimated Auto-generated benchmark | 70GB |
| meta-llama/Llama-3.3-70B-Instruct | FP16 | 85.60 tok/sEstimated Auto-generated benchmark | 137GB |
| moonshotai/Kimi-Linear-48B-A3B-Instruct | FP16 | 84.44 tok/sEstimated Auto-generated benchmark | 101GB |
| Qwen/Qwen3-32B | FP16 | 84.16 tok/sEstimated Auto-generated benchmark | 66GB |
| moonshotai/Kimi-K2-Thinking | FP16 | 83.66 tok/sEstimated Auto-generated benchmark | 1956GB |
| Qwen/Qwen2.5-32B-Instruct | FP16 | 82.94 tok/sEstimated Auto-generated benchmark | 66GB |
| mistralai/Mistral-Large-3-675B-Instruct-2512 | Q4 | 82.68 tok/sEstimated Auto-generated benchmark | 378GB |
| MiniMaxAI/MiniMax-M1-40k | Q4 | 82.65 tok/sEstimated Auto-generated benchmark | 255GB |
| deepseek-ai/deepseek-coder-33b-instruct | FP16 | 82.63 tok/sEstimated Auto-generated benchmark | 68GB |
| deepseek-ai/DeepSeek-Math-V2 | Q8 | 76.09 tok/sEstimated Auto-generated benchmark | 766GB |
| MiniMaxAI/MiniMax-VL-01 | Q4 | 75.06 tok/sEstimated Auto-generated benchmark | 256GB |
| deepseek-ai/DeepSeek-Coder-V2-Instruct-0724 | Q8 | 67.34 tok/sEstimated Auto-generated benchmark | 231GB |
| mistralai/Mixtral-8x22B-Instruct-v0.1 | FP16 | 64.33 tok/sEstimated Auto-generated benchmark | 275GB |
| mistralai/Mistral-Large-3-675B-Instruct-2512 | Q8 | 63.18 tok/sEstimated Auto-generated benchmark | 755GB |
| MiniMaxAI/MiniMax-M1-40k | Q8 | 63.16 tok/sEstimated Auto-generated benchmark | 510GB |
| MiniMaxAI/MiniMax-VL-01 | Q8 | 56.95 tok/sEstimated Auto-generated benchmark | 511GB |
| mistralai/Mistral-Large-Instruct-2411 | FP16 | 56.19 tok/sEstimated Auto-generated benchmark | 240GB |
| meta-llama/Llama-3.3-70B-Instruct | FP16 | 54.50 tok/sEstimated Auto-generated benchmark | 138GB |
| Qwen/Qwen2.5-72B-Instruct | FP16 | 54.29 tok/sEstimated Auto-generated benchmark | 141GB |
| RedHatAI/Llama-3.2-90B-Vision-Instruct-FP8-dynamic | FP16 | 53.48 tok/sEstimated Auto-generated benchmark | 176GB |
| AI-MO/Kimina-Prover-72B | FP16 | 53.30 tok/sEstimated Auto-generated benchmark | 141GB |
| Qwen/Qwen3-Next-80B-A3B-Thinking-FP8 | FP16 | 52.49 tok/sEstimated Auto-generated benchmark | 156GB |
| Qwen/Qwen3-235B-A22B | Q8 | 52.13 tok/sEstimated Auto-generated benchmark | 230GB |
| Qwen/Qwen3-Next-80B-A3B-Instruct-FP8 | FP16 | 52.03 tok/sEstimated Auto-generated benchmark | 156GB |
| Qwen/Qwen3-Next-80B-A3B-Thinking | FP16 | 51.95 tok/sEstimated Auto-generated benchmark | 156GB |
| Qwen/Qwen2.5-Math-72B-Instruct | FP16 | 51.64 tok/sEstimated Auto-generated benchmark | 142GB |
| meta-llama/Llama-3.1-70B-Instruct | FP16 | 51.56 tok/sEstimated Auto-generated benchmark | 138GB |
| Qwen/Qwen2.5-72B-Instruct | FP16 | 51.42 tok/sEstimated Auto-generated benchmark | 142GB |
| openai/gpt-oss-120b | FP16 | 50.33 tok/sEstimated Auto-generated benchmark | 235GB |
| NousResearch/Hermes-3-Llama-3.1-70B | FP16 | 49.70 tok/sEstimated Auto-generated benchmark | 138GB |
| nvidia/Llama-3.1-Nemotron-70B-Instruct-HF | FP16 | 48.82 tok/sEstimated Auto-generated benchmark | 138GB |
| Qwen/Qwen3-Next-80B-A3B-Instruct | FP16 | 48.10 tok/sEstimated Auto-generated benchmark | 156GB |
| deepseek-ai/DeepSeek-Coder-V2-Instruct-0724 | FP16 | 42.70 tok/sEstimated Auto-generated benchmark | 461GB |
| deepseek-ai/DeepSeek-Math-V2 | FP16 | 37.38 tok/sEstimated Auto-generated benchmark | 1532GB |
| Qwen/Qwen3-235B-A22B | FP16 | 32.95 tok/sEstimated Auto-generated benchmark | 460GB |
| mistralai/Mistral-Large-3-675B-Instruct-2512 | FP16 | 30.03 tok/sEstimated Auto-generated benchmark | 1509GB |
| MiniMaxAI/MiniMax-M1-40k | FP16 | 29.88 tok/sEstimated Auto-generated benchmark | 1020GB |
| MiniMaxAI/MiniMax-VL-01 | FP16 | 28.41 tok/sEstimated Auto-generated benchmark | 1021GB |
Note: Performance estimates are calculated. Real results may vary. Methodology · Submit real data
| Model | Quantization | Verdict | Estimated speed | VRAM needed |
|---|---|---|---|---|
| mistralai/Mistral-Large-3-675B-Instruct-2512 | Q4 | Not supported | 82.68 tok/sEstimated | 378GB (have 141GB) |
| mistralai/Mistral-Large-3-675B-Instruct-2512 | Q8 | Not supported | 63.18 tok/sEstimated | 755GB (have 141GB) |
| EssentialAI/rnj-1 | FP16 | Fits comfortably | 192.38 tok/sEstimated | 19GB (have 141GB) |
| EssentialAI/rnj-1 | Q8 | Fits comfortably | 365.37 tok/sEstimated | 10GB (have 141GB) |
| EssentialAI/rnj-1 | Q4 | Fits comfortably | 487.85 tok/sEstimated | 5GB (have 141GB) |
| mistralai/Mistral-Large-3-675B-Instruct-2512 | FP16 | Not supported | 30.03 tok/sEstimated | 1509GB (have 141GB) |
| lmstudio-community/Qwen3-4B-Thinking-2507-MLX-8bit | Q4 | Fits comfortably | 653.89 tok/sEstimated | 2GB (have 141GB) |
| codellama/CodeLlama-34b-hf | Q4 | Fits comfortably | 217.90 tok/sEstimated | 17GB (have 141GB) |
| unsloth/Meta-Llama-3.1-8B-Instruct | FP16 | Fits comfortably | 246.02 tok/sEstimated | 17GB (have 141GB) |
| Qwen/Qwen2.5-Coder-7B-Instruct | Q8 | Fits comfortably | 436.17 tok/sEstimated | 7GB (have 141GB) |
| Qwen/Qwen2.5-Coder-7B-Instruct | FP16 | Fits comfortably | 283.46 tok/sEstimated | 15GB (have 141GB) |
| RedHatAI/Llama-3.2-90B-Vision-Instruct-FP8-dynamic | Q4 | Fits comfortably | 124.45 tok/sEstimated | 44GB (have 141GB) |
| RedHatAI/Llama-3.2-90B-Vision-Instruct-FP8-dynamic | Q8 | Fits comfortably | 100.96 tok/sEstimated | 88GB (have 141GB) |
| lmstudio-community/Qwen3-Coder-30B-A3B-Instruct-MLX-4bit | Q8 | Fits comfortably | 264.44 tok/sEstimated | 31GB (have 141GB) |
| Qwen/Qwen2.5-Coder-7B-Instruct | Q4 | Fits comfortably | 756.47 tok/sEstimated | 4GB (have 141GB) |
| Qwen/Qwen2-1.5B-Instruct | Q8 | Fits comfortably | 472.76 tok/sEstimated | 5GB (have 141GB) |
| Qwen/Qwen3-30B-A3B-Thinking-2507 | FP16 | Fits comfortably | 157.11 tok/sEstimated | 61GB (have 141GB) |
| Qwen/Qwen3-30B-A3B-Thinking-2507 | Q8 | Fits comfortably | 251.52 tok/sEstimated | 31GB (have 141GB) |
| lmstudio-community/Qwen3-Coder-30B-A3B-Instruct-MLX-5bit | Q4 | Fits comfortably | 342.29 tok/sEstimated | 15GB (have 141GB) |
| deepseek-ai/DeepSeek-V3 | Q4 | Fits comfortably | 661.09 tok/sEstimated | 4GB (have 141GB) |
| deepseek-ai/DeepSeek-V3 | Q8 | Fits comfortably | 474.17 tok/sEstimated | 7GB (have 141GB) |
| deepseek-ai/DeepSeek-V3 | FP16 | Fits comfortably | 252.93 tok/sEstimated | 15GB (have 141GB) |
| dicta-il/dictalm2.0-instruct | Q8 | Fits comfortably | 471.20 tok/sEstimated | 7GB (have 141GB) |
| dicta-il/dictalm2.0-instruct | FP16 | Fits comfortably | 237.11 tok/sEstimated | 15GB (have 141GB) |
| Qwen/Qwen3-1.7B-Base | Q8 | Fits comfortably | 436.72 tok/sEstimated | 7GB (have 141GB) |
| Qwen/Qwen3-1.7B-Base | FP16 | Fits comfortably | 282.69 tok/sEstimated | 15GB (have 141GB) |
| ibm-granite/granite-3.3-8b-instruct | Q4 | Fits comfortably | 682.53 tok/sEstimated | 4GB (have 141GB) |
| Qwen/Qwen3-0.6B | FP16 | Fits comfortably | 259.40 tok/sEstimated | 13GB (have 141GB) |
| Gensyn/Qwen2.5-0.5B-Instruct | Q4 | Fits comfortably | 633.35 tok/sEstimated | 3GB (have 141GB) |
| google/gemma-3-1b-it | FP16 | Fits comfortably | 321.89 tok/sEstimated | 2GB (have 141GB) |
| lmstudio-community/Qwen3-4B-Thinking-2507-MLX-8bit | FP16 | Fits comfortably | 260.42 tok/sEstimated | 9GB (have 141GB) |
| Qwen/Qwen3-Next-80B-A3B-Instruct | Q8 | Fits comfortably | 103.73 tok/sEstimated | 78GB (have 141GB) |
| Qwen/Qwen3-Next-80B-A3B-Instruct | Q4 | Fits comfortably | 131.67 tok/sEstimated | 39GB (have 141GB) |
| Qwen/Qwen3-30B-A3B-Thinking-2507 | Q4 | Fits comfortably | 401.16 tok/sEstimated | 15GB (have 141GB) |
| Qwen/Qwen3-Next-80B-A3B-Instruct | FP16 | Not supported | 48.10 tok/sEstimated | 156GB (have 141GB) |
| allenai/OLMo-2-0425-1B | Q4 | Fits comfortably | 866.34 tok/sEstimated | 1GB (have 141GB) |
| Qwen/Qwen2.5-1.5B | FP16 | Fits comfortably | 250.45 tok/sEstimated | 11GB (have 141GB) |
| Qwen/Qwen2.5-14B-Instruct | Q4 | Fits comfortably | 532.29 tok/sEstimated | 7GB (have 141GB) |
| zai-org/GLM-4.6-FP8 | Q8 | Fits comfortably | 518.98 tok/sEstimated | 7GB (have 141GB) |
| zai-org/GLM-4.6-FP8 | FP16 | Fits comfortably | 266.21 tok/sEstimated | 15GB (have 141GB) |
| deepseek-ai/DeepSeek-V3.1 | FP16 | Fits comfortably | 282.24 tok/sEstimated | 15GB (have 141GB) |
| meta-llama/Llama-3.1-8B | Q4 | Fits comfortably | 729.12 tok/sEstimated | 4GB (have 141GB) |
| Qwen/Qwen3-Embedding-4B | Q4 | Fits comfortably | 637.56 tok/sEstimated | 2GB (have 141GB) |
| Qwen/Qwen3-Embedding-4B | Q8 | Fits comfortably | 527.13 tok/sEstimated | 4GB (have 141GB) |
| Qwen/Qwen3-Embedding-4B | FP16 | Fits comfortably | 254.40 tok/sEstimated | 9GB (have 141GB) |
| meta-llama/Llama-Guard-3-8B | FP16 | Fits comfortably | 280.28 tok/sEstimated | 17GB (have 141GB) |
| sshleifer/tiny-gpt2 | Q4 | Fits comfortably | 694.61 tok/sEstimated | 4GB (have 141GB) |
| sshleifer/tiny-gpt2 | Q8 | Fits comfortably | 482.01 tok/sEstimated | 7GB (have 141GB) |
| numind/NuExtract-1.5 | Q8 | Fits comfortably | 525.27 tok/sEstimated | 7GB (have 141GB) |
| lmstudio-community/DeepSeek-R1-0528-Qwen3-8B-MLX-4bit | Q4 | Fits comfortably | 654.14 tok/sEstimated | 4GB (have 141GB) |
| HuggingFaceH4/zephyr-7b-beta | Q4 | Fits comfortably | 743.73 tok/sEstimated | 4GB (have 141GB) |
| HuggingFaceH4/zephyr-7b-beta | Q8 | Fits comfortably | 448.66 tok/sEstimated | 7GB (have 141GB) |
| HuggingFaceH4/zephyr-7b-beta | FP16 | Fits comfortably | 268.78 tok/sEstimated | 15GB (have 141GB) |
| liuhaotian/llava-v1.5-7b | Q4 | Fits comfortably | 663.30 tok/sEstimated | 4GB (have 141GB) |
| liuhaotian/llava-v1.5-7b | Q8 | Fits comfortably | 505.56 tok/sEstimated | 7GB (have 141GB) |
| liuhaotian/llava-v1.5-7b | FP16 | Fits comfortably | 238.25 tok/sEstimated | 15GB (have 141GB) |
| Qwen/Qwen2.5-72B-Instruct | Q4 | Fits comfortably | 141.11 tok/sEstimated | 35GB (have 141GB) |
| Qwen/Qwen2.5-72B-Instruct | Q8 | Fits comfortably | 105.82 tok/sEstimated | 70GB (have 141GB) |
| google/gemma-3-270m-it | Q8 | Fits comfortably | 474.50 tok/sEstimated | 7GB (have 141GB) |
| google/gemma-3-270m-it | FP16 | Fits comfortably | 283.44 tok/sEstimated | 15GB (have 141GB) |
| skt/kogpt2-base-v2 | Q4 | Fits comfortably | 715.47 tok/sEstimated | 4GB (have 141GB) |
| skt/kogpt2-base-v2 | Q8 | Fits comfortably | 496.54 tok/sEstimated | 7GB (have 141GB) |
| google/gemma-2-9b-it | FP16 | Fits comfortably | 176.85 tok/sEstimated | 20GB (have 141GB) |
| Qwen/Qwen2.5-7B-Instruct | FP16 | Fits comfortably | 261.30 tok/sEstimated | 16GB (have 141GB) |
| Qwen/Qwen3-Embedding-0.6B | Q8 | Fits comfortably | 525.00 tok/sEstimated | 6GB (have 141GB) |
| Qwen/Qwen2.5-7B-Instruct | Q8 | Fits comfortably | 456.26 tok/sEstimated | 8GB (have 141GB) |
| nvidia/Llama-3.1-Nemotron-70B-Instruct-HF | Q8 | Fits comfortably | 96.56 tok/sEstimated | 69GB (have 141GB) |
| tencent/HunyuanOCR | Q4 | Fits comfortably | 828.14 tok/sEstimated | 1GB (have 141GB) |
| tencent/HunyuanOCR | Q8 | Fits comfortably | 563.83 tok/sEstimated | 2GB (have 141GB) |
| allenai/Olmo-3-7B-Think | Q4 | Fits comfortably | 649.74 tok/sEstimated | 4GB (have 141GB) |
| allenai/Olmo-3-7B-Think | Q8 | Fits comfortably | 443.05 tok/sEstimated | 8GB (have 141GB) |
| allenai/OLMo-2-0425-1B | FP16 | Fits comfortably | 331.91 tok/sEstimated | 2GB (have 141GB) |
| Qwen/Qwen3-1.7B | FP16 | Fits comfortably | 238.22 tok/sEstimated | 15GB (have 141GB) |
| Qwen/Qwen3-4B | Q4 | Fits comfortably | 710.79 tok/sEstimated | 2GB (have 141GB) |
| unsloth/Meta-Llama-3.1-8B-Instruct-bnb-4bit | Q4 | Fits comfortably | 627.76 tok/sEstimated | 4GB (have 141GB) |
| meta-llama/Llama-3.3-70B-Instruct | FP16 | Fits comfortably | 85.60 tok/sEstimated | 137GB (have 141GB) |
| Qwen/Qwen2.5-32B-Instruct | Q8 | Fits comfortably | 166.66 tok/sEstimated | 33GB (have 141GB) |
| Qwen/Qwen2.5-32B-Instruct | FP16 | Fits comfortably | 82.94 tok/sEstimated | 66GB (have 141GB) |
| mistralai/Mistral-7B-v0.1 | Q4 | Fits comfortably | 705.00 tok/sEstimated | 4GB (have 141GB) |
| mistralai/Mistral-7B-v0.1 | Q8 | Fits comfortably | 446.29 tok/sEstimated | 7GB (have 141GB) |
| mistralai/Mistral-7B-v0.1 | FP16 | Fits comfortably | 237.55 tok/sEstimated | 15GB (have 141GB) |
| Qwen/Qwen3-14B-Base | FP16 | Fits comfortably | 193.89 tok/sEstimated | 29GB (have 141GB) |
| unsloth/DeepSeek-R1-Distill-Qwen-32B-bnb-4bit | Q4 | Fits comfortably | 220.00 tok/sEstimated | 16GB (have 141GB) |
| unsloth/DeepSeek-R1-Distill-Qwen-32B-bnb-4bit | Q8 | Fits comfortably | 159.83 tok/sEstimated | 33GB (have 141GB) |
| codellama/CodeLlama-34b-hf | Q8 | Fits comfortably | 152.19 tok/sEstimated | 35GB (have 141GB) |
| codellama/CodeLlama-34b-hf | FP16 | Fits comfortably | 85.74 tok/sEstimated | 70GB (have 141GB) |
| EleutherAI/gpt-neo-125m | Q4 | Fits comfortably | 713.82 tok/sEstimated | 4GB (have 141GB) |
| EleutherAI/gpt-neo-125m | Q8 | Fits comfortably | 486.78 tok/sEstimated | 7GB (have 141GB) |
| unsloth/Llama-3.2-3B-Instruct | Q4 | Fits comfortably | 892.67 tok/sEstimated | 2GB (have 141GB) |
| unsloth/Llama-3.2-3B-Instruct | Q8 | Fits comfortably | 631.17 tok/sEstimated | 3GB (have 141GB) |
| lmstudio-community/Qwen3-4B-Thinking-2507-MLX-6bit | FP16 | Fits comfortably | 236.30 tok/sEstimated | 9GB (have 141GB) |
| RedHatAI/Meta-Llama-3.1-70B-Instruct-quantized.w4a16 | Q4 | Fits comfortably | 224.85 tok/sEstimated | 34GB (have 141GB) |
| RedHatAI/Meta-Llama-3.1-70B-Instruct-quantized.w4a16 | Q8 | Fits comfortably | 157.39 tok/sEstimated | 68GB (have 141GB) |
| RedHatAI/Meta-Llama-3.1-70B-Instruct-quantized.w4a16 | FP16 | Fits comfortably | 91.67 tok/sEstimated | 137GB (have 141GB) |
| sshleifer/tiny-gpt2 | FP16 | Fits comfortably | 278.32 tok/sEstimated | 15GB (have 141GB) |
| microsoft/Phi-3-mini-128k-instruct | Q4 | Fits comfortably | 671.72 tok/sEstimated | 4GB (have 141GB) |
| microsoft/Phi-3-mini-128k-instruct | Q8 | Fits comfortably | 529.49 tok/sEstimated | 7GB (have 141GB) |
| microsoft/Phi-3-mini-128k-instruct | FP16 | Fits comfortably | 286.11 tok/sEstimated | 15GB (have 141GB) |
| deepseek-ai/DeepSeek-V3-0324 | Q4 | Fits comfortably | 667.43 tok/sEstimated | 4GB (have 141GB) |
| Qwen/QwQ-32B-Preview | Q4 | Fits comfortably | 242.82 tok/sEstimated | 17GB (have 141GB) |
Note: Performance estimates are calculated. Real results may vary. Methodology · Submit real data
Explore how RTX 5070 stacks up for local inference workloads.
Explore how RTX 4060 Ti 16GB stacks up for local inference workloads.
Explore how RX 6800 XT stacks up for local inference workloads.
Explore how RTX 4070 Super stacks up for local inference workloads.
Explore how RTX 3080 stacks up for local inference workloads.