Loading content...
Loading GPU specifications...
Loading GPU benchmarks...
Quick Answer: NVIDIA A4000 offers 16GB VRAM and starts around $21.99. It delivers approximately 99 tokens/sec on ibm-research/PowerMoE-3b. It typically draws 140W 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 |
|---|---|---|---|
| ibm-research/PowerMoE-3b | Q4 | 99.44 tok/sEstimated Auto-generated benchmark | 2GB |
| unsloth/gemma-3-1b-it | Q4 | 99.03 tok/sEstimated Auto-generated benchmark | 1GB |
| meta-llama/Llama-Guard-3-1B | Q4 | 99.00 tok/sEstimated Auto-generated benchmark | 1GB |
| deepseek-ai/DeepSeek-OCR | Q4 | 98.25 tok/sEstimated Auto-generated benchmark | 2GB |
| meta-llama/Llama-3.2-1B | Q4 | 97.99 tok/sEstimated Auto-generated benchmark | 1GB |
| TinyLlama/TinyLlama-1.1B-Chat-v1.0 | Q4 | 97.89 tok/sEstimated Auto-generated benchmark | 1GB |
| bigcode/starcoder2-3b | Q4 | 97.75 tok/sEstimated Auto-generated benchmark | 2GB |
| google/gemma-2b | Q4 | 96.16 tok/sEstimated Auto-generated benchmark | 1GB |
| google/embeddinggemma-300m | Q4 | 95.34 tok/sEstimated Auto-generated benchmark | 1GB |
| ibm-granite/granite-3.3-2b-instruct | Q4 | 94.84 tok/sEstimated Auto-generated benchmark | 1GB |
| meta-llama/Llama-3.2-3B | Q4 | 94.78 tok/sEstimated Auto-generated benchmark | 2GB |
| LiquidAI/LFM2-1.2B | Q4 | 94.71 tok/sEstimated Auto-generated benchmark | 1GB |
| google-t5/t5-3b | Q4 | 94.48 tok/sEstimated Auto-generated benchmark | 2GB |
| Qwen/Qwen2.5-3B-Instruct | Q4 | 94.24 tok/sEstimated Auto-generated benchmark | 2GB |
| meta-llama/Llama-3.2-1B-Instruct | Q4 | 94.07 tok/sEstimated Auto-generated benchmark | 1GB |
| WeiboAI/VibeThinker-1.5B | Q4 | 94.03 tok/sEstimated Auto-generated benchmark | 1GB |
| inference-net/Schematron-3B | Q4 | 93.22 tok/sEstimated Auto-generated benchmark | 2GB |
| unsloth/Llama-3.2-3B-Instruct | Q4 | 91.93 tok/sEstimated Auto-generated benchmark | 2GB |
| meta-llama/Llama-3.2-3B-Instruct | Q4 | 90.95 tok/sEstimated Auto-generated benchmark | 2GB |
| google-bert/bert-base-uncased | Q4 | 90.95 tok/sEstimated Auto-generated benchmark | 1GB |
| facebook/sam3 | Q4 | 89.80 tok/sEstimated Auto-generated benchmark | 1GB |
| Qwen/Qwen2.5-3B | Q4 | 88.16 tok/sEstimated Auto-generated benchmark | 2GB |
| google/gemma-2-2b-it | Q4 | 87.93 tok/sEstimated Auto-generated benchmark | 1GB |
| context-labs/meta-llama-Llama-3.2-3B-Instruct-FP16 | Q4 | 86.92 tok/sEstimated Auto-generated benchmark | 2GB |
| allenai/OLMo-2-0425-1B | Q4 | 85.46 tok/sEstimated Auto-generated benchmark | 1GB |
| unsloth/Llama-3.2-1B-Instruct | Q4 | 84.74 tok/sEstimated Auto-generated benchmark | 1GB |
| meta-llama/Llama-2-7b-hf | Q4 | 84.50 tok/sEstimated Auto-generated benchmark | 4GB |
| apple/OpenELM-1_1B-Instruct | Q4 | 84.48 tok/sEstimated Auto-generated benchmark | 1GB |
| microsoft/Phi-3-mini-4k-instruct | Q4 | 84.43 tok/sEstimated Auto-generated benchmark | 4GB |
| Qwen/Qwen2.5-7B-Instruct | Q4 | 84.37 tok/sEstimated Auto-generated benchmark | 4GB |
| Qwen/Qwen2.5-Coder-7B-Instruct | Q4 | 84.24 tok/sEstimated Auto-generated benchmark | 4GB |
| Tongyi-MAI/Z-Image-Turbo | Q4 | 84.16 tok/sEstimated Auto-generated benchmark | 4GB |
| deepseek-ai/deepseek-coder-1.3b-instruct | Q4 | 84.15 tok/sEstimated Auto-generated benchmark | 2GB |
| tencent/HunyuanVideo-1.5 | Q4 | 84.10 tok/sEstimated Auto-generated benchmark | 4GB |
| Qwen/Qwen2.5-1.5B | Q4 | 84.09 tok/sEstimated Auto-generated benchmark | 3GB |
| Qwen/Qwen3-1.7B | Q4 | 84.08 tok/sEstimated Auto-generated benchmark | 4GB |
| openai-community/gpt2-xl | Q4 | 83.99 tok/sEstimated Auto-generated benchmark | 4GB |
| meta-llama/Llama-2-7b-chat-hf | Q4 | 83.93 tok/sEstimated Auto-generated benchmark | 4GB |
| Qwen/Qwen2.5-0.5B | Q4 | 83.92 tok/sEstimated Auto-generated benchmark | 3GB |
| nari-labs/Dia2-2B | Q4 | 83.68 tok/sEstimated Auto-generated benchmark | 2GB |
| Qwen/Qwen3-4B-Base | Q4 | 83.62 tok/sEstimated Auto-generated benchmark | 2GB |
| Qwen/Qwen3-1.7B-Base | Q4 | 83.59 tok/sEstimated Auto-generated benchmark | 4GB |
| tencent/HunyuanOCR | Q4 | 83.49 tok/sEstimated Auto-generated benchmark | 1GB |
| petals-team/StableBeluga2 | Q4 | 83.39 tok/sEstimated Auto-generated benchmark | 4GB |
| google/gemma-3-1b-it | Q4 | 83.34 tok/sEstimated Auto-generated benchmark | 1GB |
| swiss-ai/Apertus-8B-Instruct-2509 | Q4 | 83.33 tok/sEstimated Auto-generated benchmark | 4GB |
| unsloth/mistral-7b-v0.3-bnb-4bit | Q4 | 83.29 tok/sEstimated Auto-generated benchmark | 4GB |
| Qwen/Qwen2-1.5B-Instruct | Q4 | 83.28 tok/sEstimated Auto-generated benchmark | 3GB |
| lmstudio-community/Qwen3-4B-Thinking-2507-MLX-6bit | Q4 | 83.28 tok/sEstimated Auto-generated benchmark | 2GB |
| ibm-granite/granite-docling-258M | Q4 | 83.25 tok/sEstimated Auto-generated benchmark | 4GB |
| Qwen/Qwen2.5-1.5B-Instruct | Q4 | 83.15 tok/sEstimated Auto-generated benchmark | 3GB |
| bigscience/bloomz-560m | Q4 | 83.07 tok/sEstimated Auto-generated benchmark | 4GB |
| lmstudio-community/DeepSeek-R1-0528-Qwen3-8B-MLX-8bit | Q4 | 82.81 tok/sEstimated Auto-generated benchmark | 4GB |
| trl-internal-testing/tiny-Qwen2ForCausalLM-2.5 | Q4 | 82.77 tok/sEstimated Auto-generated benchmark | 4GB |
| Qwen/Qwen3-8B | Q4 | 82.46 tok/sEstimated Auto-generated benchmark | 4GB |
| deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct | Q4 | 82.21 tok/sEstimated Auto-generated benchmark | 4GB |
| deepseek-ai/DeepSeek-V3 | Q4 | 81.99 tok/sEstimated Auto-generated benchmark | 4GB |
| Qwen/Qwen3-8B-FP8 | Q4 | 81.99 tok/sEstimated Auto-generated benchmark | 4GB |
| Qwen/Qwen3-8B-Base | Q4 | 81.88 tok/sEstimated Auto-generated benchmark | 4GB |
| ibm-granite/granite-3.3-8b-instruct | Q4 | 81.69 tok/sEstimated Auto-generated benchmark | 4GB |
| microsoft/VibeVoice-1.5B | Q4 | 81.68 tok/sEstimated Auto-generated benchmark | 3GB |
| vikhyatk/moondream2 | Q4 | 81.57 tok/sEstimated Auto-generated benchmark | 4GB |
| deepseek-ai/DeepSeek-R1-Distill-Llama-8B | Q4 | 81.57 tok/sEstimated Auto-generated benchmark | 4GB |
| numind/NuExtract-1.5 | Q4 | 81.56 tok/sEstimated Auto-generated benchmark | 4GB |
| Qwen/Qwen2.5-0.5B-Instruct | Q4 | 81.50 tok/sEstimated Auto-generated benchmark | 3GB |
| sshleifer/tiny-gpt2 | Q4 | 81.38 tok/sEstimated Auto-generated benchmark | 4GB |
| HuggingFaceTB/SmolLM-135M | Q4 | 81.22 tok/sEstimated Auto-generated benchmark | 4GB |
| openai-community/gpt2-medium | Q4 | 81.18 tok/sEstimated Auto-generated benchmark | 4GB |
| unsloth/Meta-Llama-3.1-8B-Instruct | Q4 | 81.13 tok/sEstimated Auto-generated benchmark | 4GB |
| Qwen/Qwen3-4B-Thinking-2507 | Q4 | 80.97 tok/sEstimated Auto-generated benchmark | 2GB |
| HuggingFaceM4/tiny-random-LlamaForCausalLM | Q4 | 80.65 tok/sEstimated Auto-generated benchmark | 4GB |
| microsoft/Phi-3.5-vision-instruct | Q4 | 80.59 tok/sEstimated Auto-generated benchmark | 4GB |
| lmstudio-community/DeepSeek-R1-0528-Qwen3-8B-MLX-4bit | Q4 | 80.41 tok/sEstimated Auto-generated benchmark | 4GB |
| llamafactory/tiny-random-Llama-3 | Q4 | 80.31 tok/sEstimated Auto-generated benchmark | 4GB |
| Qwen/Qwen3-Embedding-8B | Q4 | 80.26 tok/sEstimated Auto-generated benchmark | 4GB |
| Qwen/Qwen2.5-7B | Q4 | 80.22 tok/sEstimated Auto-generated benchmark | 4GB |
| Qwen/Qwen2.5-Math-1.5B | Q4 | 80.17 tok/sEstimated Auto-generated benchmark | 3GB |
| HuggingFaceH4/zephyr-7b-beta | Q4 | 80.16 tok/sEstimated Auto-generated benchmark | 4GB |
| openai-community/gpt2 | Q4 | 79.99 tok/sEstimated Auto-generated benchmark | 4GB |
| facebook/opt-125m | Q4 | 79.93 tok/sEstimated Auto-generated benchmark | 4GB |
| microsoft/Phi-3-mini-128k-instruct | Q4 | 79.88 tok/sEstimated Auto-generated benchmark | 4GB |
| dicta-il/dictalm2.0-instruct | Q4 | 79.74 tok/sEstimated Auto-generated benchmark | 4GB |
| hmellor/tiny-random-LlamaForCausalLM | Q4 | 79.64 tok/sEstimated Auto-generated benchmark | 4GB |
| lmstudio-community/Qwen3-4B-Thinking-2507-MLX-4bit | Q4 | 79.63 tok/sEstimated Auto-generated benchmark | 2GB |
| parler-tts/parler-tts-large-v1 | Q4 | 79.48 tok/sEstimated Auto-generated benchmark | 4GB |
| NousResearch/Meta-Llama-3.1-8B-Instruct | Q4 | 79.47 tok/sEstimated Auto-generated benchmark | 4GB |
| lmstudio-community/Qwen3-4B-Thinking-2507-MLX-8bit | Q4 | 79.41 tok/sEstimated Auto-generated benchmark | 2GB |
| deepseek-ai/DeepSeek-R1 | Q4 | 79.32 tok/sEstimated Auto-generated benchmark | 4GB |
| meta-llama/Llama-3.2-3B-Instruct | Q4 | 79.18 tok/sEstimated Auto-generated benchmark | 2GB |
| unsloth/Meta-Llama-3.1-8B-Instruct-bnb-4bit | Q4 | 79.07 tok/sEstimated Auto-generated benchmark | 4GB |
| IlyaGusev/saiga_llama3_8b | Q4 | 78.79 tok/sEstimated Auto-generated benchmark | 4GB |
| Qwen/Qwen2.5-Coder-1.5B | Q4 | 78.62 tok/sEstimated Auto-generated benchmark | 3GB |
| Qwen/Qwen3-Reranker-0.6B | Q4 | 78.59 tok/sEstimated Auto-generated benchmark | 3GB |
| microsoft/DialoGPT-medium | Q4 | 78.51 tok/sEstimated Auto-generated benchmark | 4GB |
| rinna/japanese-gpt-neox-small | Q4 | 78.51 tok/sEstimated Auto-generated benchmark | 4GB |
| distilbert/distilgpt2 | Q4 | 78.25 tok/sEstimated Auto-generated benchmark | 4GB |
| Qwen/Qwen3-4B-Thinking-2507-FP8 | Q4 | 78.24 tok/sEstimated Auto-generated benchmark | 2GB |
| meta-llama/Meta-Llama-3-8B-Instruct | Q4 | 78.10 tok/sEstimated Auto-generated benchmark | 4GB |
| MiniMaxAI/MiniMax-M2 | Q4 | 78.00 tok/sEstimated Auto-generated benchmark | 4GB |
| openai-community/gpt2-large | Q4 | 77.78 tok/sEstimated Auto-generated benchmark | 4GB |
| black-forest-labs/FLUX.2-dev | Q4 | 77.75 tok/sEstimated Auto-generated benchmark | 4GB |
| google/gemma-3-270m-it | Q4 | 77.12 tok/sEstimated Auto-generated benchmark | 4GB |
| deepseek-ai/DeepSeek-V3-0324 | Q4 | 77.10 tok/sEstimated Auto-generated benchmark | 4GB |
| skt/kogpt2-base-v2 | Q4 | 77.08 tok/sEstimated Auto-generated benchmark | 4GB |
| GSAI-ML/LLaDA-8B-Instruct | Q4 | 77.01 tok/sEstimated Auto-generated benchmark | 4GB |
| microsoft/DialoGPT-small | Q4 | 77.01 tok/sEstimated Auto-generated benchmark | 4GB |
| Qwen/Qwen2.5-7B-Instruct | Q4 | 76.98 tok/sEstimated Auto-generated benchmark | 4GB |
| mistralai/Mistral-7B-Instruct-v0.1 | Q4 | 76.87 tok/sEstimated Auto-generated benchmark | 4GB |
| microsoft/phi-4 | Q4 | 76.83 tok/sEstimated Auto-generated benchmark | 4GB |
| black-forest-labs/FLUX.1-dev | Q4 | 76.75 tok/sEstimated Auto-generated benchmark | 4GB |
| microsoft/Phi-3.5-mini-instruct | Q4 | 76.28 tok/sEstimated Auto-generated benchmark | 4GB |
| meta-llama/Llama-Guard-3-8B | Q4 | 76.12 tok/sEstimated Auto-generated benchmark | 4GB |
| Qwen/Qwen2-0.5B-Instruct | Q4 | 76.02 tok/sEstimated Auto-generated benchmark | 3GB |
| mistralai/Mistral-7B-Instruct-v0.2 | Q4 | 75.75 tok/sEstimated Auto-generated benchmark | 4GB |
| deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B | Q4 | 75.50 tok/sEstimated Auto-generated benchmark | 3GB |
| meta-llama/Llama-3.1-8B | Q4 | 75.49 tok/sEstimated Auto-generated benchmark | 4GB |
| huggyllama/llama-7b | Q4 | 75.39 tok/sEstimated Auto-generated benchmark | 4GB |
| deepseek-ai/DeepSeek-R1-0528 | Q4 | 75.30 tok/sEstimated Auto-generated benchmark | 4GB |
| rednote-hilab/dots.ocr | Q4 | 75.23 tok/sEstimated Auto-generated benchmark | 4GB |
| microsoft/phi-2 | Q4 | 75.21 tok/sEstimated Auto-generated benchmark | 4GB |
| GSAI-ML/LLaDA-8B-Base | Q4 | 75.14 tok/sEstimated Auto-generated benchmark | 4GB |
| microsoft/Phi-4-multimodal-instruct | Q4 | 74.84 tok/sEstimated Auto-generated benchmark | 4GB |
| EleutherAI/pythia-70m-deduped | Q4 | 74.74 tok/sEstimated Auto-generated benchmark | 4GB |
| lmsys/vicuna-7b-v1.5 | Q4 | 74.54 tok/sEstimated Auto-generated benchmark | 4GB |
| Qwen/Qwen-Image-Edit-2509 | Q4 | 74.48 tok/sEstimated Auto-generated benchmark | 4GB |
| meta-llama/Llama-3.1-8B-Instruct | Q4 | 73.80 tok/sEstimated Auto-generated benchmark | 4GB |
| microsoft/Phi-3.5-mini-instruct | Q4 | 73.40 tok/sEstimated Auto-generated benchmark | 2GB |
| Qwen/Qwen2-0.5B | Q4 | 73.39 tok/sEstimated Auto-generated benchmark | 3GB |
| HuggingFaceTB/SmolLM2-135M | Q4 | 72.88 tok/sEstimated Auto-generated benchmark | 4GB |
| Gensyn/Qwen2.5-0.5B-Instruct | Q4 | 72.44 tok/sEstimated Auto-generated benchmark | 3GB |
| Qwen/Qwen3-4B-Instruct-2507 | Q4 | 72.36 tok/sEstimated Auto-generated benchmark | 2GB |
| Qwen/Qwen3-0.6B | Q4 | 72.01 tok/sEstimated Auto-generated benchmark | 3GB |
| Alibaba-NLP/gte-Qwen2-1.5B-instruct | Q4 | 71.91 tok/sEstimated Auto-generated benchmark | 3GB |
| trl-internal-testing/tiny-LlamaForCausalLM-3.2 | Q4 | 71.91 tok/sEstimated Auto-generated benchmark | 4GB |
| deepseek-ai/DeepSeek-R1-Distill-Qwen-7B | Q4 | 71.83 tok/sEstimated Auto-generated benchmark | 4GB |
| trl-internal-testing/tiny-random-LlamaForCausalLM | Q4 | 71.81 tok/sEstimated Auto-generated benchmark | 4GB |
| BSC-LT/salamandraTA-7b-instruct | Q4 | 71.64 tok/sEstimated Auto-generated benchmark | 4GB |
| liuhaotian/llava-v1.5-7b | Q4 | 71.15 tok/sEstimated Auto-generated benchmark | 4GB |
| meta-llama/Meta-Llama-3-8B | Q4 | 71.09 tok/sEstimated Auto-generated benchmark | 4GB |
| deepseek-ai/DeepSeek-V3.1 | Q4 | 71.05 tok/sEstimated Auto-generated benchmark | 4GB |
| Qwen/Qwen2-7B-Instruct | Q4 | 70.85 tok/sEstimated Auto-generated benchmark | 4GB |
| Qwen/Qwen3-Embedding-4B | Q4 | 70.78 tok/sEstimated Auto-generated benchmark | 2GB |
| Qwen/Qwen3-4B | Q4 | 70.64 tok/sEstimated Auto-generated benchmark | 2GB |
| zai-org/GLM-4.6-FP8 | Q4 | 70.64 tok/sEstimated Auto-generated benchmark | 4GB |
| google-bert/bert-base-uncased | Q8 | 70.33 tok/sEstimated Auto-generated benchmark | 1GB |
| Qwen/Qwen3-Embedding-0.6B | Q4 | 70.09 tok/sEstimated Auto-generated benchmark | 3GB |
| microsoft/Phi-4-mini-instruct | Q4 | 70.08 tok/sEstimated Auto-generated benchmark | 4GB |
| Qwen/Qwen3-0.6B-Base | Q4 | 69.95 tok/sEstimated Auto-generated benchmark | 3GB |
| kaitchup/Phi-3-mini-4k-instruct-gptq-4bit | Q4 | 69.95 tok/sEstimated Auto-generated benchmark | 2GB |
| mistralai/Mistral-7B-v0.1 | Q4 | 69.91 tok/sEstimated Auto-generated benchmark | 4GB |
| zai-org/GLM-4.5-Air | Q4 | 69.69 tok/sEstimated Auto-generated benchmark | 4GB |
| EleutherAI/gpt-neo-125m | Q4 | 69.48 tok/sEstimated Auto-generated benchmark | 4GB |
| context-labs/meta-llama-Llama-3.2-3B-Instruct-FP16 | Q8 | 69.31 tok/sEstimated Auto-generated benchmark | 3GB |
| allenai/Olmo-3-7B-Think | Q4 | 69.25 tok/sEstimated Auto-generated benchmark | 4GB |
| Qwen/Qwen2.5-3B-Instruct | Q8 | 68.92 tok/sEstimated Auto-generated benchmark | 3GB |
| TinyLlama/TinyLlama-1.1B-Chat-v1.0 | Q8 | 68.25 tok/sEstimated Auto-generated benchmark | 1GB |
| meta-llama/Llama-3.2-3B-Instruct | Q8 | 68.17 tok/sEstimated Auto-generated benchmark | 3GB |
| deepseek-ai/deepseek-coder-1.3b-instruct | Q8 | 67.99 tok/sEstimated Auto-generated benchmark | 3GB |
| WeiboAI/VibeThinker-1.5B | Q8 | 67.79 tok/sEstimated Auto-generated benchmark | 2GB |
| google/gemma-2b | Q8 | 67.69 tok/sEstimated Auto-generated benchmark | 2GB |
| google/gemma-3-1b-it | Q8 | 67.59 tok/sEstimated Auto-generated benchmark | 1GB |
| Qwen/Qwen2.5-3B | Q8 | 67.49 tok/sEstimated Auto-generated benchmark | 3GB |
| facebook/sam3 | Q8 | 66.98 tok/sEstimated Auto-generated benchmark | 1GB |
| unsloth/Llama-3.2-1B-Instruct | Q8 | 66.97 tok/sEstimated Auto-generated benchmark | 1GB |
| meta-llama/Llama-3.2-3B | Q8 | 65.78 tok/sEstimated Auto-generated benchmark | 3GB |
| meta-llama/Llama-3.2-1B | Q8 | 65.47 tok/sEstimated Auto-generated benchmark | 1GB |
| deepseek-ai/DeepSeek-OCR | Q8 | 64.81 tok/sEstimated Auto-generated benchmark | 4GB |
| allenai/OLMo-2-0425-1B | Q8 | 64.26 tok/sEstimated Auto-generated benchmark | 1GB |
| google/embeddinggemma-300m | Q8 | 64.02 tok/sEstimated Auto-generated benchmark | 1GB |
| ai-forever/ruGPT-3.5-13B | Q4 | 62.88 tok/sEstimated Auto-generated benchmark | 7GB |
| google/gemma-2-9b-it | Q4 | 62.81 tok/sEstimated Auto-generated benchmark | 5GB |
| unsloth/gemma-3-1b-it | Q8 | 62.65 tok/sEstimated Auto-generated benchmark | 1GB |
| mistralai/Ministral-3-14B-Instruct-2512 | Q4 | 62.47 tok/sEstimated Auto-generated benchmark | 8GB |
| meta-llama/Llama-3.2-1B-Instruct | Q8 | 62.40 tok/sEstimated Auto-generated benchmark | 1GB |
| LiquidAI/LFM2-1.2B | Q8 | 62.35 tok/sEstimated Auto-generated benchmark | 2GB |
| nari-labs/Dia2-2B | Q8 | 62.30 tok/sEstimated Auto-generated benchmark | 3GB |
| Qwen/Qwen3-14B | Q4 | 61.69 tok/sEstimated Auto-generated benchmark | 7GB |
| ibm-research/PowerMoE-3b | Q8 | 61.67 tok/sEstimated Auto-generated benchmark | 3GB |
| EssentialAI/rnj-1 | Q4 | 61.17 tok/sEstimated Auto-generated benchmark | 5GB |
| ibm-granite/granite-3.3-2b-instruct | Q8 | 61.12 tok/sEstimated Auto-generated benchmark | 2GB |
| tencent/HunyuanOCR | Q8 | 60.43 tok/sEstimated Auto-generated benchmark | 2GB |
| NousResearch/Hermes-3-Llama-3.1-8B | Q4 | 60.38 tok/sEstimated Auto-generated benchmark | 4GB |
| bigcode/starcoder2-3b | Q8 | 60.34 tok/sEstimated Auto-generated benchmark | 3GB |
| Qwen/Qwen2.5-14B-Instruct | Q4 | 59.76 tok/sEstimated Auto-generated benchmark | 8GB |
| inference-net/Schematron-3B | Q8 | 59.47 tok/sEstimated Auto-generated benchmark | 3GB |
| meta-llama/Llama-Guard-3-1B | Q8 | 59.40 tok/sEstimated Auto-generated benchmark | 1GB |
| nvidia/NVIDIA-Nemotron-Nano-9B-v2 | Q4 | 59.26 tok/sEstimated Auto-generated benchmark | 5GB |
| OpenPipe/Qwen3-14B-Instruct | Q4 | 59.24 tok/sEstimated Auto-generated benchmark | 7GB |
| apple/OpenELM-1_1B-Instruct | Q8 | 59.21 tok/sEstimated Auto-generated benchmark | 1GB |
| unsloth/Meta-Llama-3.1-8B-Instruct | Q8 | 59.14 tok/sEstimated Auto-generated benchmark | 9GB |
| google-t5/t5-3b | Q8 | 59.08 tok/sEstimated Auto-generated benchmark | 3GB |
| microsoft/Phi-3-medium-128k-instruct | Q4 | 59.00 tok/sEstimated Auto-generated benchmark | 7GB |
| lmstudio-community/DeepSeek-R1-0528-Qwen3-8B-MLX-4bit | Q8 | 58.90 tok/sEstimated Auto-generated benchmark | 9GB |
| google/gemma-2-2b-it | Q8 | 58.85 tok/sEstimated Auto-generated benchmark | 2GB |
| Qwen/Qwen2.5-1.5B | Q8 | 58.83 tok/sEstimated Auto-generated benchmark | 5GB |
| deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct | Q8 | 58.79 tok/sEstimated Auto-generated benchmark | 7GB |
| unsloth/Llama-3.2-3B-Instruct | Q8 | 58.72 tok/sEstimated Auto-generated benchmark | 3GB |
| mistralai/Mistral-7B-v0.1 | Q8 | 58.60 tok/sEstimated Auto-generated benchmark | 7GB |
| GSAI-ML/LLaDA-8B-Instruct | Q8 | 58.54 tok/sEstimated Auto-generated benchmark | 9GB |
| HuggingFaceH4/zephyr-7b-beta | Q8 | 58.36 tok/sEstimated Auto-generated benchmark | 7GB |
| black-forest-labs/FLUX.1-dev | Q8 | 58.33 tok/sEstimated Auto-generated benchmark | 8GB |
| microsoft/Phi-3.5-vision-instruct | Q8 | 58.22 tok/sEstimated Auto-generated benchmark | 7GB |
| Qwen/Qwen2-0.5B | Q8 | 58.22 tok/sEstimated Auto-generated benchmark | 5GB |
| meta-llama/Llama-2-7b-hf | Q8 | 58.06 tok/sEstimated Auto-generated benchmark | 7GB |
| deepseek-ai/DeepSeek-R1 | Q8 | 57.95 tok/sEstimated Auto-generated benchmark | 7GB |
| vikhyatk/moondream2 | Q8 | 57.86 tok/sEstimated Auto-generated benchmark | 7GB |
| rednote-hilab/dots.ocr | Q8 | 57.77 tok/sEstimated Auto-generated benchmark | 7GB |
| rinna/japanese-gpt-neox-small | Q8 | 57.74 tok/sEstimated Auto-generated benchmark | 7GB |
| meta-llama/Llama-Guard-3-8B | Q8 | 57.66 tok/sEstimated Auto-generated benchmark | 9GB |
| facebook/opt-125m | Q8 | 57.60 tok/sEstimated Auto-generated benchmark | 7GB |
| Qwen/Qwen3-8B-FP8 | Q8 | 57.52 tok/sEstimated Auto-generated benchmark | 9GB |
| microsoft/phi-4 | Q8 | 57.39 tok/sEstimated Auto-generated benchmark | 7GB |
| kaitchup/Phi-3-mini-4k-instruct-gptq-4bit | Q8 | 57.33 tok/sEstimated Auto-generated benchmark | 4GB |
| GSAI-ML/LLaDA-8B-Base | Q8 | 57.31 tok/sEstimated Auto-generated benchmark | 9GB |
| huggyllama/llama-7b | Q8 | 57.14 tok/sEstimated Auto-generated benchmark | 7GB |
| meta-llama/Meta-Llama-3-8B | Q8 | 57.09 tok/sEstimated Auto-generated benchmark | 9GB |
| deepseek-ai/DeepSeek-R1-0528 | Q8 | 57.02 tok/sEstimated Auto-generated benchmark | 7GB |
| lmstudio-community/DeepSeek-R1-0528-Qwen3-8B-MLX-8bit | Q8 | 56.97 tok/sEstimated Auto-generated benchmark | 9GB |
| petals-team/StableBeluga2 | Q8 | 56.88 tok/sEstimated Auto-generated benchmark | 7GB |
| microsoft/VibeVoice-1.5B | Q8 | 56.80 tok/sEstimated Auto-generated benchmark | 5GB |
| deepseek-ai/DeepSeek-R1-Distill-Llama-8B | Q8 | 56.78 tok/sEstimated Auto-generated benchmark | 9GB |
| allenai/Olmo-3-7B-Think | Q8 | 56.66 tok/sEstimated Auto-generated benchmark | 8GB |
| Qwen/Qwen3-4B | Q8 | 56.63 tok/sEstimated Auto-generated benchmark | 4GB |
| dicta-il/dictalm2.0-instruct | Q8 | 56.45 tok/sEstimated Auto-generated benchmark | 7GB |
| openai-community/gpt2-xl | Q8 | 56.38 tok/sEstimated Auto-generated benchmark | 7GB |
| Qwen/Qwen3-8B | Q8 | 56.24 tok/sEstimated Auto-generated benchmark | 9GB |
| Qwen/Qwen2.5-Coder-1.5B | Q8 | 56.22 tok/sEstimated Auto-generated benchmark | 5GB |
| trl-internal-testing/tiny-LlamaForCausalLM-3.2 | Q8 | 56.16 tok/sEstimated Auto-generated benchmark | 7GB |
| NousResearch/Meta-Llama-3.1-8B-Instruct | Q8 | 56.05 tok/sEstimated Auto-generated benchmark | 9GB |
| Alibaba-NLP/gte-Qwen2-1.5B-instruct | Q8 | 56.03 tok/sEstimated Auto-generated benchmark | 5GB |
| numind/NuExtract-1.5 | Q8 | 56.02 tok/sEstimated Auto-generated benchmark | 7GB |
| Qwen/Qwen3-4B-Instruct-2507 | Q8 | 55.99 tok/sEstimated Auto-generated benchmark | 4GB |
| meta-llama/Llama-2-13b-chat-hf | Q4 | 55.90 tok/sEstimated Auto-generated benchmark | 7GB |
| google/gemma-3-270m-it | Q8 | 55.78 tok/sEstimated Auto-generated benchmark | 7GB |
| deepseek-ai/DeepSeek-V3-0324 | Q8 | 55.64 tok/sEstimated Auto-generated benchmark | 7GB |
| skt/kogpt2-base-v2 | Q8 | 55.55 tok/sEstimated Auto-generated benchmark | 7GB |
| openai-community/gpt2 | Q8 | 55.51 tok/sEstimated Auto-generated benchmark | 7GB |
| Qwen/Qwen3-1.7B-Base | Q8 | 55.46 tok/sEstimated Auto-generated benchmark | 7GB |
| Qwen/Qwen3-Embedding-8B | Q8 | 55.29 tok/sEstimated Auto-generated benchmark | 9GB |
| meta-llama/Llama-3.1-8B | Q8 | 55.25 tok/sEstimated Auto-generated benchmark | 9GB |
| deepseek-ai/DeepSeek-V3.1 | Q8 | 55.10 tok/sEstimated Auto-generated benchmark | 7GB |
| Qwen/Qwen2-7B-Instruct | Q8 | 55.00 tok/sEstimated Auto-generated benchmark | 7GB |
| IlyaGusev/saiga_llama3_8b | Q8 | 54.76 tok/sEstimated Auto-generated benchmark | 9GB |
| Qwen/Qwen3-0.6B | Q8 | 54.74 tok/sEstimated Auto-generated benchmark | 6GB |
| microsoft/phi-2 | Q8 | 54.73 tok/sEstimated Auto-generated benchmark | 7GB |
| deepseek-ai/DeepSeek-R1-Distill-Qwen-7B | Q8 | 54.72 tok/sEstimated Auto-generated benchmark | 7GB |
| Qwen/Qwen3-1.7B | Q8 | 54.62 tok/sEstimated Auto-generated benchmark | 7GB |
| EleutherAI/gpt-neo-125m | Q8 | 54.60 tok/sEstimated Auto-generated benchmark | 7GB |
| Qwen/Qwen2.5-0.5B-Instruct | Q8 | 54.16 tok/sEstimated Auto-generated benchmark | 5GB |
| lmsys/vicuna-7b-v1.5 | Q8 | 54.14 tok/sEstimated Auto-generated benchmark | 7GB |
| Qwen/Qwen2.5-7B | Q8 | 54.10 tok/sEstimated Auto-generated benchmark | 7GB |
| MiniMaxAI/MiniMax-M2 | Q8 | 54.02 tok/sEstimated Auto-generated benchmark | 7GB |
| Qwen/Qwen3-Embedding-0.6B | Q8 | 53.99 tok/sEstimated Auto-generated benchmark | 6GB |
| Qwen/Qwen3-Reranker-0.6B | Q8 | 53.99 tok/sEstimated Auto-generated benchmark | 6GB |
| Qwen/Qwen3-4B-Base | Q8 | 53.99 tok/sEstimated Auto-generated benchmark | 4GB |
| microsoft/DialoGPT-small | Q8 | 53.81 tok/sEstimated Auto-generated benchmark | 7GB |
| openai-community/gpt2-large | Q8 | 53.74 tok/sEstimated Auto-generated benchmark | 7GB |
| Qwen/Qwen2.5-14B | Q4 | 53.55 tok/sEstimated Auto-generated benchmark | 7GB |
| meta-llama/Meta-Llama-3-8B-Instruct | Q8 | 53.44 tok/sEstimated Auto-generated benchmark | 9GB |
| trl-internal-testing/tiny-random-LlamaForCausalLM | Q8 | 53.31 tok/sEstimated Auto-generated benchmark | 7GB |
| EleutherAI/pythia-70m-deduped | Q8 | 53.31 tok/sEstimated Auto-generated benchmark | 7GB |
| mistralai/Mistral-7B-Instruct-v0.2 | Q8 | 53.22 tok/sEstimated Auto-generated benchmark | 7GB |
| Qwen/Qwen3-14B-Base | Q4 | 53.02 tok/sEstimated Auto-generated benchmark | 7GB |
| meta-llama/Llama-3.2-3B-Instruct | Q8 | 53.00 tok/sEstimated Auto-generated benchmark | 3GB |
| microsoft/Phi-4-multimodal-instruct | Q8 | 52.75 tok/sEstimated Auto-generated benchmark | 7GB |
| Qwen/Qwen2.5-14B-Instruct | Q4 | 52.72 tok/sEstimated Auto-generated benchmark | 7GB |
| meta-llama/Llama-3.1-8B-Instruct | Q4 | 52.68 tok/sEstimated Auto-generated benchmark | 4GB |
| unsloth/Meta-Llama-3.1-8B-Instruct-bnb-4bit | Q8 | 52.59 tok/sEstimated Auto-generated benchmark | 9GB |
| Qwen/Qwen2-0.5B-Instruct | Q8 | 52.46 tok/sEstimated Auto-generated benchmark | 5GB |
| distilbert/distilgpt2 | Q8 | 52.42 tok/sEstimated Auto-generated benchmark | 7GB |
| zai-org/GLM-4.6-FP8 | Q8 | 52.39 tok/sEstimated Auto-generated benchmark | 7GB |
| Qwen/Qwen2.5-Math-1.5B | Q8 | 52.31 tok/sEstimated Auto-generated benchmark | 5GB |
| HuggingFaceTB/SmolLM2-135M | Q8 | 52.04 tok/sEstimated Auto-generated benchmark | 7GB |
| lmstudio-community/Qwen3-4B-Thinking-2507-MLX-6bit | Q8 | 51.99 tok/sEstimated Auto-generated benchmark | 4GB |
| liuhaotian/llava-v1.5-7b | Q8 | 51.92 tok/sEstimated Auto-generated benchmark | 7GB |
| lmstudio-community/Qwen3-4B-Thinking-2507-MLX-8bit | Q8 | 51.88 tok/sEstimated Auto-generated benchmark | 4GB |
| swiss-ai/Apertus-8B-Instruct-2509 | Q8 | 51.84 tok/sEstimated Auto-generated benchmark | 9GB |
| hmellor/tiny-random-LlamaForCausalLM | Q8 | 51.83 tok/sEstimated Auto-generated benchmark | 7GB |
| Qwen/Qwen-Image-Edit-2509 | Q8 | 51.78 tok/sEstimated Auto-generated benchmark | 8GB |
| zai-org/GLM-4.5-Air | Q8 | 51.68 tok/sEstimated Auto-generated benchmark | 7GB |
| mistralai/Mistral-7B-Instruct-v0.1 | Q8 | 51.63 tok/sEstimated Auto-generated benchmark | 7GB |
| tencent/HunyuanVideo-1.5 | Q8 | 51.59 tok/sEstimated Auto-generated benchmark | 8GB |
| HuggingFaceM4/tiny-random-LlamaForCausalLM | Q8 | 51.56 tok/sEstimated Auto-generated benchmark | 7GB |
| Qwen/Qwen3-4B-Thinking-2507 | Q8 | 51.43 tok/sEstimated Auto-generated benchmark | 4GB |
| unsloth/mistral-7b-v0.3-bnb-4bit | Q8 | 51.29 tok/sEstimated Auto-generated benchmark | 7GB |
| microsoft/Phi-3.5-mini-instruct | Q8 | 51.24 tok/sEstimated Auto-generated benchmark | 7GB |
| llamafactory/tiny-random-Llama-3 | Q8 | 51.20 tok/sEstimated Auto-generated benchmark | 7GB |
| ibm-granite/granite-3.3-8b-instruct | Q8 | 51.06 tok/sEstimated Auto-generated benchmark | 9GB |
| microsoft/DialoGPT-medium | Q8 | 50.98 tok/sEstimated Auto-generated benchmark | 7GB |
| Qwen/Qwen3-Embedding-4B | Q8 | 50.93 tok/sEstimated Auto-generated benchmark | 4GB |
| HuggingFaceTB/SmolLM-135M | Q8 | 50.92 tok/sEstimated Auto-generated benchmark | 7GB |
| ibm-granite/granite-docling-258M | Q8 | 50.81 tok/sEstimated Auto-generated benchmark | 7GB |
| microsoft/Phi-3.5-mini-instruct | Q8 | 50.81 tok/sEstimated Auto-generated benchmark | 4GB |
| parler-tts/parler-tts-large-v1 | Q8 | 50.80 tok/sEstimated Auto-generated benchmark | 7GB |
| Qwen/Qwen3-0.6B-Base | Q8 | 50.75 tok/sEstimated Auto-generated benchmark | 6GB |
| microsoft/Phi-4-mini-instruct | Q8 | 50.68 tok/sEstimated Auto-generated benchmark | 7GB |
| sshleifer/tiny-gpt2 | Q8 | 50.61 tok/sEstimated Auto-generated benchmark | 7GB |
| Qwen/Qwen3-8B-Base | Q8 | 50.61 tok/sEstimated Auto-generated benchmark | 9GB |
| trl-internal-testing/tiny-Qwen2ForCausalLM-2.5 | Q8 | 50.50 tok/sEstimated Auto-generated benchmark | 7GB |
| lmstudio-community/Qwen3-4B-Thinking-2507-MLX-4bit | Q8 | 50.16 tok/sEstimated Auto-generated benchmark | 4GB |
| BSC-LT/salamandraTA-7b-instruct | Q8 | 49.82 tok/sEstimated Auto-generated benchmark | 7GB |
| Qwen/Qwen2.5-0.5B | Q8 | 49.74 tok/sEstimated Auto-generated benchmark | 5GB |
| meta-llama/Llama-2-7b-chat-hf | Q8 | 49.67 tok/sEstimated Auto-generated benchmark | 7GB |
| Tongyi-MAI/Z-Image-Turbo | Q8 | 49.61 tok/sEstimated Auto-generated benchmark | 8GB |
| Qwen/Qwen2.5-1.5B-Instruct | Q8 | 49.54 tok/sEstimated Auto-generated benchmark | 5GB |
| Qwen/Qwen2.5-7B-Instruct | Q8 | 49.48 tok/sEstimated Auto-generated benchmark | 7GB |
| Qwen/Qwen2.5-Coder-7B-Instruct | Q8 | 49.46 tok/sEstimated Auto-generated benchmark | 7GB |
| Gensyn/Qwen2.5-0.5B-Instruct | Q8 | 49.35 tok/sEstimated Auto-generated benchmark | 5GB |
| deepseek-ai/DeepSeek-V3 | Q8 | 49.29 tok/sEstimated Auto-generated benchmark | 7GB |
| microsoft/Phi-3-mini-128k-instruct | Q8 | 49.18 tok/sEstimated Auto-generated benchmark | 7GB |
| deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B | Q8 | 49.04 tok/sEstimated Auto-generated benchmark | 5GB |
| Qwen/Qwen3-4B-Thinking-2507-FP8 | Q8 | 49.00 tok/sEstimated Auto-generated benchmark | 4GB |
| meta-llama/Llama-3.1-8B-Instruct | Q8 | 48.99 tok/sEstimated Auto-generated benchmark | 9GB |
| Qwen/Qwen2-1.5B-Instruct | Q8 | 48.98 tok/sEstimated Auto-generated benchmark | 5GB |
| Qwen/Qwen2.5-7B-Instruct | Q8 | 48.96 tok/sEstimated Auto-generated benchmark | 8GB |
| black-forest-labs/FLUX.2-dev | Q8 | 48.95 tok/sEstimated Auto-generated benchmark | 8GB |
| microsoft/Phi-3-mini-4k-instruct | Q8 | 48.78 tok/sEstimated Auto-generated benchmark | 7GB |
| openai-community/gpt2-medium | Q8 | 48.61 tok/sEstimated Auto-generated benchmark | 7GB |
| bigscience/bloomz-560m | Q8 | 48.51 tok/sEstimated Auto-generated benchmark | 7GB |
| unsloth/gpt-oss-20b-unsloth-bnb-4bit | Q4 | 45.40 tok/sEstimated Auto-generated benchmark | 10GB |
| Qwen/Qwen3-30B-A3B-Thinking-2507 | Q4 | 44.84 tok/sEstimated Auto-generated benchmark | 15GB |
| lmstudio-community/Qwen3-Coder-30B-A3B-Instruct-MLX-4bit | Q4 | 44.79 tok/sEstimated Auto-generated benchmark | 15GB |
| Qwen/Qwen3-30B-A3B-Instruct-2507-FP8 | Q4 | 44.11 tok/sEstimated Auto-generated benchmark | 15GB |
| nvidia/NVIDIA-Nemotron-Nano-9B-v2 | Q8 | 43.87 tok/sEstimated Auto-generated benchmark | 10GB |
| Qwen/Qwen3-30B-A3B | Q4 | 43.86 tok/sEstimated Auto-generated benchmark | 15GB |
| meta-llama/Llama-3.1-8B-Instruct | Q8 | 43.83 tok/sEstimated Auto-generated benchmark | 9GB |
| mlx-community/gpt-oss-20b-MXFP4-Q8 | Q4 | 42.76 tok/sEstimated Auto-generated benchmark | 10GB |
| microsoft/Phi-3-medium-128k-instruct | Q8 | 42.06 tok/sEstimated Auto-generated benchmark | 14GB |
| lmstudio-community/Qwen3-Coder-30B-A3B-Instruct-MLX-6bit | Q4 | 41.72 tok/sEstimated Auto-generated benchmark | 15GB |
| mistralai/Ministral-3-14B-Instruct-2512 | Q8 | 41.66 tok/sEstimated Auto-generated benchmark | 16GB |
| OpenPipe/Qwen3-14B-Instruct | Q8 | 41.43 tok/sEstimated Auto-generated benchmark | 14GB |
| meta-llama/Llama-2-13b-chat-hf | Q8 | 41.38 tok/sEstimated Auto-generated benchmark | 13GB |
| openai/gpt-oss-safeguard-20b | Q4 | 41.36 tok/sEstimated Auto-generated benchmark | 11GB |
| lmstudio-community/Qwen3-Coder-30B-A3B-Instruct-MLX-5bit | Q4 | 41.03 tok/sEstimated Auto-generated benchmark | 15GB |
| Qwen/Qwen3-14B | Q8 | 40.74 tok/sEstimated Auto-generated benchmark | 14GB |
| openai/gpt-oss-20b | Q4 | 40.31 tok/sEstimated Auto-generated benchmark | 10GB |
| Qwen/Qwen3-Coder-30B-A3B-Instruct | Q4 | 40.19 tok/sEstimated Auto-generated benchmark | 15GB |
| EssentialAI/rnj-1 | Q8 | 40.13 tok/sEstimated Auto-generated benchmark | 10GB |
| unsloth/gpt-oss-20b-BF16 | Q4 | 39.94 tok/sEstimated Auto-generated benchmark | 10GB |
| google/gemma-2-27b-it | Q4 | 39.64 tok/sEstimated Auto-generated benchmark | 14GB |
| google/gemma-2-9b-it | Q8 | 39.24 tok/sEstimated Auto-generated benchmark | 10GB |
| NousResearch/Hermes-3-Llama-3.1-8B | Q8 | 39.14 tok/sEstimated Auto-generated benchmark | 9GB |
| Qwen/Qwen2.5-14B-Instruct | Q8 | 38.89 tok/sEstimated Auto-generated benchmark | 15GB |
| Qwen/Qwen3-30B-A3B-Instruct-2507 | Q4 | 38.62 tok/sEstimated Auto-generated benchmark | 15GB |
| ai-forever/ruGPT-3.5-13B | Q8 | 38.61 tok/sEstimated Auto-generated benchmark | 13GB |
| lmstudio-community/Qwen3-Coder-30B-A3B-Instruct-MLX-8bit | Q4 | 38.50 tok/sEstimated Auto-generated benchmark | 15GB |
| ibm-granite/granite-3.3-2b-instruct | FP16 | 38.35 tok/sEstimated Auto-generated benchmark | 4GB |
| apple/OpenELM-1_1B-Instruct | FP16 | 38.34 tok/sEstimated Auto-generated benchmark | 2GB |
| inference-net/Schematron-3B | FP16 | 38.29 tok/sEstimated Auto-generated benchmark | 6GB |
| ibm-research/PowerMoE-3b | FP16 | 38.26 tok/sEstimated Auto-generated benchmark | 6GB |
| mistralai/Mistral-Small-Instruct-2409 | Q4 | 38.19 tok/sEstimated Auto-generated benchmark | 11GB |
| google-t5/t5-3b | FP16 | 37.80 tok/sEstimated Auto-generated benchmark | 6GB |
| google/gemma-2b | FP16 | 37.77 tok/sEstimated Auto-generated benchmark | 4GB |
| google/embeddinggemma-300m | FP16 | 37.73 tok/sEstimated Auto-generated benchmark | 1GB |
| Qwen/Qwen3-14B-Base | Q8 | 37.55 tok/sEstimated Auto-generated benchmark | 14GB |
| facebook/sam3 | FP16 | 37.48 tok/sEstimated Auto-generated benchmark | 2GB |
| meta-llama/Llama-Guard-3-1B | FP16 | 37.27 tok/sEstimated Auto-generated benchmark | 2GB |
| unsloth/Llama-3.2-3B-Instruct | FP16 | 37.17 tok/sEstimated Auto-generated benchmark | 6GB |
| Qwen/Qwen2.5-14B | Q8 | 36.88 tok/sEstimated Auto-generated benchmark | 14GB |
| Qwen/Qwen2.5-14B-Instruct | Q8 | 36.66 tok/sEstimated Auto-generated benchmark | 14GB |
| unsloth/Llama-3.2-1B-Instruct | FP16 | 35.83 tok/sEstimated Auto-generated benchmark | 2GB |
| TinyLlama/TinyLlama-1.1B-Chat-v1.0 | FP16 | 35.49 tok/sEstimated Auto-generated benchmark | 2GB |
| context-labs/meta-llama-Llama-3.2-3B-Instruct-FP16 | FP16 | 35.17 tok/sEstimated Auto-generated benchmark | 6GB |
| meta-llama/Llama-3.2-1B-Instruct | FP16 | 35.16 tok/sEstimated Auto-generated benchmark | 2GB |
| tencent/HunyuanOCR | FP16 | 35.13 tok/sEstimated Auto-generated benchmark | 3GB |
| deepseek-ai/DeepSeek-OCR | FP16 | 35.09 tok/sEstimated Auto-generated benchmark | 7GB |
| bigcode/starcoder2-3b | FP16 | 35.02 tok/sEstimated Auto-generated benchmark | 6GB |
| google/gemma-2-2b-it | FP16 | 34.73 tok/sEstimated Auto-generated benchmark | 4GB |
| unsloth/gemma-3-1b-it | FP16 | 34.73 tok/sEstimated Auto-generated benchmark | 2GB |
| WeiboAI/VibeThinker-1.5B | FP16 | 34.58 tok/sEstimated Auto-generated benchmark | 4GB |
| google-bert/bert-base-uncased | FP16 | 34.05 tok/sEstimated Auto-generated benchmark | 1GB |
| nari-labs/Dia2-2B | FP16 | 33.75 tok/sEstimated Auto-generated benchmark | 5GB |
| Qwen/Qwen2.5-3B-Instruct | FP16 | 33.73 tok/sEstimated Auto-generated benchmark | 6GB |
| meta-llama/Llama-3.2-1B | FP16 | 33.64 tok/sEstimated Auto-generated benchmark | 2GB |
| LiquidAI/LFM2-1.2B | FP16 | 32.94 tok/sEstimated Auto-generated benchmark | 4GB |
| deepseek-ai/deepseek-coder-1.3b-instruct | FP16 | 32.83 tok/sEstimated Auto-generated benchmark | 6GB |
| google/gemma-3-1b-it | FP16 | 32.78 tok/sEstimated Auto-generated benchmark | 2GB |
| meta-llama/Llama-3.2-3B | FP16 | 32.71 tok/sEstimated Auto-generated benchmark | 6GB |
| unsloth/gpt-oss-20b-unsloth-bnb-4bit | Q8 | 32.49 tok/sEstimated Auto-generated benchmark | 20GB |
| Qwen/Qwen2.5-3B | FP16 | 32.42 tok/sEstimated Auto-generated benchmark | 6GB |
| allenai/OLMo-2-0425-1B | FP16 | 32.35 tok/sEstimated Auto-generated benchmark | 2GB |
| Qwen/Qwen3-30B-A3B-Instruct-2507-FP8 | Q8 | 32.31 tok/sEstimated Auto-generated benchmark | 31GB |
| meta-llama/Llama-3.2-3B-Instruct | FP16 | 32.26 tok/sEstimated Auto-generated benchmark | 6GB |
| lmstudio-community/Qwen3-Coder-30B-A3B-Instruct-MLX-6bit | Q8 | 32.19 tok/sEstimated Auto-generated benchmark | 31GB |
| lmstudio-community/Qwen3-Coder-30B-A3B-Instruct-MLX-8bit | Q8 | 32.13 tok/sEstimated Auto-generated benchmark | 31GB |
| Qwen/Qwen2.5-Math-1.5B | FP16 | 32.05 tok/sEstimated Auto-generated benchmark | 11GB |
| lmstudio-community/Qwen3-4B-Thinking-2507-MLX-8bit | FP16 | 32.00 tok/sEstimated Auto-generated benchmark | 9GB |
| lmstudio-community/Qwen3-Coder-30B-A3B-Instruct-MLX-4bit | Q8 | 31.98 tok/sEstimated Auto-generated benchmark | 31GB |
| Gensyn/Qwen2.5-0.5B-Instruct | FP16 | 31.94 tok/sEstimated Auto-generated benchmark | 11GB |
| lmstudio-community/Qwen3-4B-Thinking-2507-MLX-6bit | FP16 | 31.93 tok/sEstimated Auto-generated benchmark | 9GB |
| Qwen/Qwen3-Embedding-0.6B | FP16 | 31.90 tok/sEstimated Auto-generated benchmark | 13GB |
| dicta-il/dictalm2.0-instruct | FP16 | 31.82 tok/sEstimated Auto-generated benchmark | 15GB |
| Qwen/Qwen3-4B-Thinking-2507-FP8 | FP16 | 31.73 tok/sEstimated Auto-generated benchmark | 9GB |
| HuggingFaceTB/SmolLM2-135M | FP16 | 31.54 tok/sEstimated Auto-generated benchmark | 15GB |
| deepseek-ai/DeepSeek-R1-Distill-Llama-8B | FP16 | 31.52 tok/sEstimated Auto-generated benchmark | 17GB |
| microsoft/Phi-3-mini-4k-instruct | FP16 | 31.52 tok/sEstimated Auto-generated benchmark | 15GB |
| lmstudio-community/Qwen3-Coder-30B-A3B-Instruct-MLX-5bit | Q8 | 31.46 tok/sEstimated Auto-generated benchmark | 31GB |
| vikhyatk/moondream2 | FP16 | 31.46 tok/sEstimated Auto-generated benchmark | 15GB |
| sshleifer/tiny-gpt2 | FP16 | 31.46 tok/sEstimated Auto-generated benchmark | 15GB |
| lmstudio-community/DeepSeek-R1-0528-Qwen3-8B-MLX-4bit | FP16 | 31.44 tok/sEstimated Auto-generated benchmark | 17GB |
| google/gemma-3-270m-it | FP16 | 31.44 tok/sEstimated Auto-generated benchmark | 15GB |
| mistralai/Mistral-7B-Instruct-v0.2 | FP16 | 31.42 tok/sEstimated Auto-generated benchmark | 15GB |
| microsoft/phi-2 | FP16 | 31.31 tok/sEstimated Auto-generated benchmark | 15GB |
| NousResearch/Meta-Llama-3.1-8B-Instruct | FP16 | 31.17 tok/sEstimated Auto-generated benchmark | 17GB |
| mistralai/Mistral-Small-Instruct-2409 | Q8 | 31.15 tok/sEstimated Auto-generated benchmark | 23GB |
| Qwen/Qwen3-4B-Base | FP16 | 31.12 tok/sEstimated Auto-generated benchmark | 9GB |
| numind/NuExtract-1.5 | FP16 | 31.08 tok/sEstimated Auto-generated benchmark | 15GB |
| skt/kogpt2-base-v2 | FP16 | 31.02 tok/sEstimated Auto-generated benchmark | 15GB |
| mlx-community/gpt-oss-20b-MXFP4-Q8 | Q8 | 30.92 tok/sEstimated Auto-generated benchmark | 20GB |
| trl-internal-testing/tiny-random-LlamaForCausalLM | FP16 | 30.87 tok/sEstimated Auto-generated benchmark | 15GB |
| microsoft/DialoGPT-medium | FP16 | 30.75 tok/sEstimated Auto-generated benchmark | 15GB |
| lmsys/vicuna-7b-v1.5 | FP16 | 30.63 tok/sEstimated Auto-generated benchmark | 15GB |
| MiniMaxAI/MiniMax-M2 | FP16 | 30.62 tok/sEstimated Auto-generated benchmark | 15GB |
| Qwen/Qwen3-0.6B | FP16 | 30.61 tok/sEstimated Auto-generated benchmark | 13GB |
| openai/gpt-oss-safeguard-20b | Q8 | 30.58 tok/sEstimated Auto-generated benchmark | 22GB |
| Qwen/Qwen2.5-0.5B-Instruct | FP16 | 30.51 tok/sEstimated Auto-generated benchmark | 11GB |
| distilbert/distilgpt2 | FP16 | 30.48 tok/sEstimated Auto-generated benchmark | 15GB |
| trl-internal-testing/tiny-LlamaForCausalLM-3.2 | FP16 | 30.45 tok/sEstimated Auto-generated benchmark | 15GB |
| HuggingFaceM4/tiny-random-LlamaForCausalLM | FP16 | 30.42 tok/sEstimated Auto-generated benchmark | 15GB |
| Qwen/Qwen2-0.5B | FP16 | 30.37 tok/sEstimated Auto-generated benchmark | 11GB |
| Qwen/Qwen3-8B | FP16 | 30.33 tok/sEstimated Auto-generated benchmark | 17GB |
| openai-community/gpt2-xl | FP16 | 30.29 tok/sEstimated Auto-generated benchmark | 15GB |
| Alibaba-NLP/gte-Qwen2-1.5B-instruct | FP16 | 30.23 tok/sEstimated Auto-generated benchmark | 11GB |
| deepseek-ai/DeepSeek-R1-0528 | FP16 | 30.20 tok/sEstimated Auto-generated benchmark | 15GB |
| microsoft/Phi-3-mini-128k-instruct | FP16 | 30.16 tok/sEstimated Auto-generated benchmark | 15GB |
| unsloth/Meta-Llama-3.1-8B-Instruct-bnb-4bit | FP16 | 30.14 tok/sEstimated Auto-generated benchmark | 17GB |
| ibm-granite/granite-3.3-8b-instruct | FP16 | 30.07 tok/sEstimated Auto-generated benchmark | 17GB |
| Qwen/Qwen2.5-7B-Instruct | FP16 | 30.05 tok/sEstimated Auto-generated benchmark | 15GB |
| tencent/HunyuanVideo-1.5 | FP16 | 30.03 tok/sEstimated Auto-generated benchmark | 16GB |
| meta-llama/Llama-3.1-8B | FP16 | 30.03 tok/sEstimated Auto-generated benchmark | 17GB |
| microsoft/Phi-3.5-vision-instruct | FP16 | 29.98 tok/sEstimated Auto-generated benchmark | 15GB |
| deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B | FP16 | 29.91 tok/sEstimated Auto-generated benchmark | 11GB |
| huggyllama/llama-7b | FP16 | 29.91 tok/sEstimated Auto-generated benchmark | 15GB |
| lmstudio-community/Qwen3-4B-Thinking-2507-MLX-4bit | FP16 | 29.90 tok/sEstimated Auto-generated benchmark | 9GB |
| Qwen/Qwen2-1.5B-Instruct | FP16 | 29.59 tok/sEstimated Auto-generated benchmark | 11GB |
| Qwen/Qwen3-0.6B-Base | FP16 | 29.56 tok/sEstimated Auto-generated benchmark | 13GB |
| Qwen/Qwen3-30B-A3B | Q8 | 29.56 tok/sEstimated Auto-generated benchmark | 31GB |
| baichuan-inc/Baichuan-M2-32B | Q4 | 29.42 tok/sEstimated Auto-generated benchmark | 16GB |
| google/gemma-2-27b-it | Q8 | 29.38 tok/sEstimated Auto-generated benchmark | 28GB |
| microsoft/Phi-3.5-mini-instruct | FP16 | 29.34 tok/sEstimated Auto-generated benchmark | 15GB |
| microsoft/phi-4 | FP16 | 29.14 tok/sEstimated Auto-generated benchmark | 15GB |
| openai-community/gpt2-large | FP16 | 29.13 tok/sEstimated Auto-generated benchmark | 15GB |
| zai-org/GLM-4.6-FP8 | FP16 | 29.11 tok/sEstimated Auto-generated benchmark | 15GB |
| kaitchup/Phi-3-mini-4k-instruct-gptq-4bit | FP16 | 29.03 tok/sEstimated Auto-generated benchmark | 9GB |
| facebook/opt-125m | FP16 | 29.00 tok/sEstimated Auto-generated benchmark | 15GB |
| meta-llama/Llama-Guard-3-8B | FP16 | 29.00 tok/sEstimated Auto-generated benchmark | 17GB |
| meta-llama/Meta-Llama-3-70B-Instruct | Q4 | 28.98 tok/sEstimated Auto-generated benchmark | 34GB |
| Qwen/Qwen3-Coder-30B-A3B-Instruct | Q8 | 28.97 tok/sEstimated Auto-generated benchmark | 31GB |
| Qwen/Qwen2.5-7B | FP16 | 28.96 tok/sEstimated Auto-generated benchmark | 15GB |
| ibm-granite/granite-docling-258M | FP16 | 28.94 tok/sEstimated Auto-generated benchmark | 15GB |
| meta-llama/Llama-3.1-70B-Instruct | Q4 | 28.93 tok/sEstimated Auto-generated benchmark | 34GB |
| Qwen/Qwen3-4B-Instruct-2507 | FP16 | 28.89 tok/sEstimated Auto-generated benchmark | 9GB |
| Qwen/Qwen2.5-1.5B | FP16 | 28.85 tok/sEstimated Auto-generated benchmark | 11GB |
| Qwen/Qwen3-Embedding-8B | FP16 | 28.84 tok/sEstimated Auto-generated benchmark | 17GB |
| Qwen/Qwen2.5-1.5B-Instruct | FP16 | 28.79 tok/sEstimated Auto-generated benchmark | 11GB |
| codellama/CodeLlama-34b-hf | Q4 | 28.64 tok/sEstimated Auto-generated benchmark | 17GB |
| microsoft/DialoGPT-small | FP16 | 28.60 tok/sEstimated Auto-generated benchmark | 15GB |
| llamafactory/tiny-random-Llama-3 | FP16 | 28.59 tok/sEstimated Auto-generated benchmark | 15GB |
| Qwen/Qwen3-1.7B | FP16 | 28.53 tok/sEstimated Auto-generated benchmark | 15GB |
| liuhaotian/llava-v1.5-7b | FP16 | 28.53 tok/sEstimated Auto-generated benchmark | 15GB |
| Qwen/Qwen3-Embedding-4B | FP16 | 28.49 tok/sEstimated Auto-generated benchmark | 9GB |
| unsloth/gpt-oss-20b-BF16 | Q8 | 28.49 tok/sEstimated Auto-generated benchmark | 20GB |
| lmstudio-community/DeepSeek-R1-0528-Qwen3-8B-MLX-8bit | FP16 | 28.46 tok/sEstimated Auto-generated benchmark | 17GB |
| Qwen/Qwen3-Reranker-0.6B | FP16 | 28.46 tok/sEstimated Auto-generated benchmark | 13GB |
| Tongyi-MAI/Z-Image-Turbo | FP16 | 28.45 tok/sEstimated Auto-generated benchmark | 16GB |
| Qwen/Qwen3-1.7B-Base | FP16 | 28.41 tok/sEstimated Auto-generated benchmark | 15GB |
| meta-llama/Llama-2-7b-hf | FP16 | 28.33 tok/sEstimated Auto-generated benchmark | 15GB |
| bigscience/bloomz-560m | FP16 | 28.30 tok/sEstimated Auto-generated benchmark | 15GB |
| hmellor/tiny-random-LlamaForCausalLM | FP16 | 28.20 tok/sEstimated Auto-generated benchmark | 15GB |
| swiss-ai/Apertus-8B-Instruct-2509 | FP16 | 28.19 tok/sEstimated Auto-generated benchmark | 17GB |
| meta-llama/Meta-Llama-3-8B | FP16 | 28.13 tok/sEstimated Auto-generated benchmark | 17GB |
| Qwen/Qwen3-8B-Base | FP16 | 28.09 tok/sEstimated Auto-generated benchmark | 17GB |
| Qwen/Qwen2.5-7B-Instruct | FP16 | 28.07 tok/sEstimated Auto-generated benchmark | 16GB |
| meta-llama/Llama-3.1-8B-Instruct | FP16 | 28.02 tok/sEstimated Auto-generated benchmark | 17GB |
| HuggingFaceTB/SmolLM-135M | FP16 | 27.97 tok/sEstimated Auto-generated benchmark | 15GB |
| deepseek-ai/DeepSeek-V3-0324 | FP16 | 27.92 tok/sEstimated Auto-generated benchmark | 15GB |
| Qwen/Qwen2.5-Coder-7B-Instruct | FP16 | 27.85 tok/sEstimated Auto-generated benchmark | 15GB |
| meta-llama/Meta-Llama-3-8B-Instruct | FP16 | 27.75 tok/sEstimated Auto-generated benchmark | 17GB |
| deepseek-ai/DeepSeek-R1-Distill-Qwen-7B | FP16 | 27.73 tok/sEstimated Auto-generated benchmark | 15GB |
| microsoft/VibeVoice-1.5B | FP16 | 27.70 tok/sEstimated Auto-generated benchmark | 11GB |
| microsoft/Phi-4-multimodal-instruct | FP16 | 27.70 tok/sEstimated Auto-generated benchmark | 15GB |
| deepseek-ai/DeepSeek-V3 | FP16 | 27.67 tok/sEstimated Auto-generated benchmark | 15GB |
| Qwen/Qwen3-4B-Thinking-2507 | FP16 | 27.58 tok/sEstimated Auto-generated benchmark | 9GB |
| meta-llama/Llama-3.2-3B-Instruct | FP16 | 27.57 tok/sEstimated Auto-generated benchmark | 7GB |
| unsloth/Meta-Llama-3.1-8B-Instruct | FP16 | 27.53 tok/sEstimated Auto-generated benchmark | 17GB |
| Qwen/Qwen3-4B | FP16 | 27.47 tok/sEstimated Auto-generated benchmark | 9GB |
| deepseek-ai/DeepSeek-V3.1 | FP16 | 27.44 tok/sEstimated Auto-generated benchmark | 15GB |
| petals-team/StableBeluga2 | FP16 | 27.42 tok/sEstimated Auto-generated benchmark | 15GB |
| rednote-hilab/dots.ocr | FP16 | 27.41 tok/sEstimated Auto-generated benchmark | 15GB |
| Qwen/Qwen3-30B-A3B-Instruct-2507 | Q8 | 27.39 tok/sEstimated Auto-generated benchmark | 31GB |
| Qwen/Qwen3-30B-A3B-Thinking-2507 | Q8 | 27.39 tok/sEstimated Auto-generated benchmark | 31GB |
| rinna/japanese-gpt-neox-small | FP16 | 27.34 tok/sEstimated Auto-generated benchmark | 15GB |
| Qwen/Qwen3-8B-FP8 | FP16 | 27.32 tok/sEstimated Auto-generated benchmark | 17GB |
| Qwen/Qwen2.5-32B | Q4 | 27.31 tok/sEstimated Auto-generated benchmark | 16GB |
| parler-tts/parler-tts-large-v1 | FP16 | 27.27 tok/sEstimated Auto-generated benchmark | 15GB |
| BSC-LT/salamandraTA-7b-instruct | FP16 | 27.26 tok/sEstimated Auto-generated benchmark | 15GB |
| EleutherAI/pythia-70m-deduped | FP16 | 27.26 tok/sEstimated Auto-generated benchmark | 15GB |
| microsoft/Phi-3.5-mini-instruct | FP16 | 27.23 tok/sEstimated Auto-generated benchmark | 8GB |
| Qwen/Qwen2.5-0.5B | FP16 | 27.16 tok/sEstimated Auto-generated benchmark | 11GB |
| openai/gpt-oss-20b | Q8 | 27.14 tok/sEstimated Auto-generated benchmark | 20GB |
| GSAI-ML/LLaDA-8B-Instruct | FP16 | 27.12 tok/sEstimated Auto-generated benchmark | 17GB |
| trl-internal-testing/tiny-Qwen2ForCausalLM-2.5 | FP16 | 27.12 tok/sEstimated Auto-generated benchmark | 15GB |
| 01-ai/Yi-1.5-34B-Chat | Q4 | 27.07 tok/sEstimated Auto-generated benchmark | 18GB |
| Qwen/Qwen2.5-Coder-1.5B | FP16 | 27.05 tok/sEstimated Auto-generated benchmark | 11GB |
| deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct | FP16 | 27.03 tok/sEstimated Auto-generated benchmark | 15GB |
| Qwen/Qwen2-7B-Instruct | FP16 | 26.99 tok/sEstimated Auto-generated benchmark | 15GB |
| IlyaGusev/saiga_llama3_8b | FP16 | 26.99 tok/sEstimated Auto-generated benchmark | 17GB |
| moonshotai/Kimi-K2-Thinking | Q4 | 26.97 tok/sEstimated Auto-generated benchmark | 489GB |
| EleutherAI/gpt-neo-125m | FP16 | 26.96 tok/sEstimated Auto-generated benchmark | 15GB |
| HuggingFaceH4/zephyr-7b-beta | FP16 | 26.96 tok/sEstimated Auto-generated benchmark | 15GB |
| black-forest-labs/FLUX.2-dev | FP16 | 26.92 tok/sEstimated Auto-generated benchmark | 16GB |
| Qwen/Qwen2-0.5B-Instruct | FP16 | 26.92 tok/sEstimated Auto-generated benchmark | 11GB |
| black-forest-labs/FLUX.1-dev | FP16 | 26.92 tok/sEstimated Auto-generated benchmark | 16GB |
| Qwen/Qwen2.5-Coder-32B-Instruct | Q4 | 26.91 tok/sEstimated Auto-generated benchmark | 17GB |
| GSAI-ML/LLaDA-8B-Base | FP16 | 26.76 tok/sEstimated Auto-generated benchmark | 17GB |
| deepseek-ai/DeepSeek-R1 | FP16 | 26.72 tok/sEstimated Auto-generated benchmark | 15GB |
| microsoft/Phi-4-mini-instruct | FP16 | 26.69 tok/sEstimated Auto-generated benchmark | 15GB |
| openai-community/gpt2 | FP16 | 26.69 tok/sEstimated Auto-generated benchmark | 15GB |
| Qwen/Qwen-Image-Edit-2509 | FP16 | 26.68 tok/sEstimated Auto-generated benchmark | 16GB |
| unsloth/mistral-7b-v0.3-bnb-4bit | FP16 | 26.68 tok/sEstimated Auto-generated benchmark | 15GB |
| meta-llama/Llama-2-7b-chat-hf | FP16 | 26.65 tok/sEstimated Auto-generated benchmark | 15GB |
| Qwen/QwQ-32B-Preview | Q4 | 26.63 tok/sEstimated Auto-generated benchmark | 17GB |
| zai-org/GLM-4.5-Air | FP16 | 26.63 tok/sEstimated Auto-generated benchmark | 15GB |
| moonshotai/Kimi-Linear-48B-A3B-Instruct | Q4 | 26.57 tok/sEstimated Auto-generated benchmark | 25GB |
| mistralai/Mistral-7B-Instruct-v0.1 | FP16 | 26.50 tok/sEstimated Auto-generated benchmark | 15GB |
| allenai/Olmo-3-7B-Think | FP16 | 26.43 tok/sEstimated Auto-generated benchmark | 16GB |
| openai-community/gpt2-medium | FP16 | 26.39 tok/sEstimated Auto-generated benchmark | 15GB |
| mistralai/Mistral-7B-v0.1 | FP16 | 26.32 tok/sEstimated Auto-generated benchmark | 15GB |
| Qwen/Qwen2.5-32B-Instruct | Q4 | 25.76 tok/sEstimated Auto-generated benchmark | 17GB |
| meta-llama/Llama-3.3-70B-Instruct | Q4 | 25.69 tok/sEstimated Auto-generated benchmark | 34GB |
| deepseek-ai/DeepSeek-V2.5 | Q4 | 24.69 tok/sEstimated Auto-generated benchmark | 328GB |
| dphn/dolphin-2.9.1-yi-1.5-34b | Q4 | 24.68 tok/sEstimated Auto-generated benchmark | 17GB |
| Qwen/Qwen3-32B | Q4 | 24.62 tok/sEstimated Auto-generated benchmark | 16GB |
| deepseek-ai/DeepSeek-R1-Distill-Qwen-32B | Q4 | 24.59 tok/sEstimated Auto-generated benchmark | 16GB |
| RedHatAI/Meta-Llama-3.1-70B-Instruct-quantized.w4a16 | Q4 | 24.49 tok/sEstimated Auto-generated benchmark | 34GB |
| unsloth/DeepSeek-R1-Distill-Qwen-32B-bnb-4bit | Q4 | 24.37 tok/sEstimated Auto-generated benchmark | 16GB |
| RedHatAI/Llama-3.3-70B-Instruct-FP8-dynamic | Q4 | 24.34 tok/sEstimated Auto-generated benchmark | 34GB |
| Qwen/Qwen2.5-32B-Instruct | Q4 | 24.32 tok/sEstimated Auto-generated benchmark | 16GB |
| deepseek-ai/deepseek-coder-33b-instruct | Q4 | 24.32 tok/sEstimated Auto-generated benchmark | 17GB |
| meta-llama/Llama-3.1-8B-Instruct | FP16 | 24.06 tok/sEstimated Auto-generated benchmark | 17GB |
| meta-llama/Llama-2-13b-chat-hf | FP16 | 23.95 tok/sEstimated Auto-generated benchmark | 27GB |
| nvidia/NVIDIA-Nemotron-Nano-9B-v2 | FP16 | 23.79 tok/sEstimated Auto-generated benchmark | 19GB |
| microsoft/Phi-3-medium-128k-instruct | FP16 | 23.58 tok/sEstimated Auto-generated benchmark | 29GB |
| OpenPipe/Qwen3-14B-Instruct | FP16 | 23.55 tok/sEstimated Auto-generated benchmark | 29GB |
| EssentialAI/rnj-1 | FP16 | 23.44 tok/sEstimated Auto-generated benchmark | 19GB |
| NousResearch/Hermes-3-Llama-3.1-8B | FP16 | 23.14 tok/sEstimated Auto-generated benchmark | 17GB |
| Qwen/Qwen3-14B | FP16 | 23.03 tok/sEstimated Auto-generated benchmark | 29GB |
| Qwen/Qwen2.5-14B | FP16 | 22.86 tok/sEstimated Auto-generated benchmark | 29GB |
| Qwen/Qwen3-14B-Base | FP16 | 22.64 tok/sEstimated Auto-generated benchmark | 29GB |
| mistralai/Ministral-3-14B-Instruct-2512 | FP16 | 22.49 tok/sEstimated Auto-generated benchmark | 32GB |
| Qwen/Qwen2.5-14B-Instruct | FP16 | 22.08 tok/sEstimated Auto-generated benchmark | 30GB |
| Qwen/Qwen2.5-14B-Instruct | FP16 | 21.65 tok/sEstimated Auto-generated benchmark | 29GB |
| ai-forever/ruGPT-3.5-13B | FP16 | 21.62 tok/sEstimated Auto-generated benchmark | 27GB |
| mistralai/Mixtral-8x22B-Instruct-v0.1 | Q4 | 20.74 tok/sEstimated Auto-generated benchmark | 69GB |
| Qwen/Qwen2.5-32B-Instruct | Q8 | 20.68 tok/sEstimated Auto-generated benchmark | 33GB |
| google/gemma-2-9b-it | FP16 | 20.45 tok/sEstimated Auto-generated benchmark | 20GB |
| Qwen/Qwen2.5-Coder-32B-Instruct | Q8 | 20.33 tok/sEstimated Auto-generated benchmark | 34GB |
| RedHatAI/Llama-3.3-70B-Instruct-FP8-dynamic | Q8 | 20.27 tok/sEstimated Auto-generated benchmark | 68GB |
| Qwen/Qwen3-32B | Q8 | 20.18 tok/sEstimated Auto-generated benchmark | 33GB |
| 01-ai/Yi-1.5-34B-Chat | Q8 | 20.17 tok/sEstimated Auto-generated benchmark | 35GB |
| deepseek-ai/deepseek-coder-33b-instruct | Q8 | 20.12 tok/sEstimated Auto-generated benchmark | 34GB |
| meta-llama/Llama-3.3-70B-Instruct | Q8 | 19.28 tok/sEstimated Auto-generated benchmark | 68GB |
| deepseek-ai/DeepSeek-V2.5 | Q8 | 19.20 tok/sEstimated Auto-generated benchmark | 656GB |
| Qwen/QwQ-32B-Preview | Q8 | 18.97 tok/sEstimated Auto-generated benchmark | 34GB |
| baichuan-inc/Baichuan-M2-32B | Q8 | 18.92 tok/sEstimated Auto-generated benchmark | 33GB |
| dphn/dolphin-2.9.1-yi-1.5-34b | Q8 | 18.91 tok/sEstimated Auto-generated benchmark | 35GB |
| moonshotai/Kimi-Linear-48B-A3B-Instruct | Q8 | 18.49 tok/sEstimated Auto-generated benchmark | 50GB |
| moonshotai/Kimi-K2-Thinking | Q8 | 18.43 tok/sEstimated Auto-generated benchmark | 978GB |
| deepseek-ai/DeepSeek-R1-Distill-Qwen-32B | Q8 | 18.30 tok/sEstimated Auto-generated benchmark | 33GB |
| meta-llama/Llama-3.1-70B-Instruct | Q8 | 17.98 tok/sEstimated Auto-generated benchmark | 68GB |
| Qwen/Qwen2.5-32B | Q8 | 17.94 tok/sEstimated Auto-generated benchmark | 33GB |
| codellama/CodeLlama-34b-hf | Q8 | 17.72 tok/sEstimated Auto-generated benchmark | 35GB |
| Qwen/Qwen3-30B-A3B-Instruct-2507 | FP16 | 17.57 tok/sEstimated Auto-generated benchmark | 61GB |
| unsloth/DeepSeek-R1-Distill-Qwen-32B-bnb-4bit | Q8 | 17.50 tok/sEstimated Auto-generated benchmark | 33GB |
| meta-llama/Meta-Llama-3-70B-Instruct | Q8 | 17.48 tok/sEstimated Auto-generated benchmark | 68GB |
| RedHatAI/Meta-Llama-3.1-70B-Instruct-quantized.w4a16 | Q8 | 17.27 tok/sEstimated Auto-generated benchmark | 68GB |
| lmstudio-community/Qwen3-Coder-30B-A3B-Instruct-MLX-4bit | FP16 | 17.18 tok/sEstimated Auto-generated benchmark | 61GB |
| google/gemma-2-27b-it | FP16 | 17.12 tok/sEstimated Auto-generated benchmark | 56GB |
| Qwen/Qwen2.5-32B-Instruct | Q8 | 17.04 tok/sEstimated Auto-generated benchmark | 34GB |
| Qwen/Qwen3-Next-80B-A3B-Instruct | Q4 | 16.91 tok/sEstimated Auto-generated benchmark | 39GB |
| Qwen/Qwen2.5-72B-Instruct | Q4 | 16.89 tok/sEstimated Auto-generated benchmark | 35GB |
| lmstudio-community/Qwen3-Coder-30B-A3B-Instruct-MLX-6bit | FP16 | 16.86 tok/sEstimated Auto-generated benchmark | 61GB |
| meta-llama/Llama-3.1-70B-Instruct | Q4 | 16.70 tok/sEstimated Auto-generated benchmark | 34GB |
| openai/gpt-oss-safeguard-20b | FP16 | 16.67 tok/sEstimated Auto-generated benchmark | 44GB |
| Qwen/Qwen3-Coder-30B-A3B-Instruct | FP16 | 16.47 tok/sEstimated Auto-generated benchmark | 61GB |
| openai/gpt-oss-20b | FP16 | 16.38 tok/sEstimated Auto-generated benchmark | 41GB |
| mistralai/Mistral-Large-Instruct-2411 | Q4 | 16.31 tok/sEstimated Auto-generated benchmark | 60GB |
| lmstudio-community/Qwen3-Coder-30B-A3B-Instruct-MLX-8bit | FP16 | 16.24 tok/sEstimated Auto-generated benchmark | 61GB |
| nvidia/Llama-3.1-Nemotron-70B-Instruct-HF | Q4 | 15.78 tok/sEstimated Auto-generated benchmark | 34GB |
| AI-MO/Kimina-Prover-72B | Q4 | 15.74 tok/sEstimated Auto-generated benchmark | 35GB |
| mistralai/Mistral-Small-Instruct-2409 | FP16 | 15.71 tok/sEstimated Auto-generated benchmark | 46GB |
| unsloth/gpt-oss-20b-BF16 | FP16 | 15.64 tok/sEstimated Auto-generated benchmark | 41GB |
| Qwen/Qwen3-30B-A3B-Thinking-2507 | FP16 | 15.58 tok/sEstimated Auto-generated benchmark | 61GB |
| RedHatAI/Llama-3.2-90B-Vision-Instruct-FP8-dynamic | Q4 | 15.44 tok/sEstimated Auto-generated benchmark | 44GB |
| Qwen/Qwen3-Next-80B-A3B-Thinking | Q4 | 15.37 tok/sEstimated Auto-generated benchmark | 39GB |
| mlx-community/gpt-oss-20b-MXFP4-Q8 | FP16 | 15.14 tok/sEstimated Auto-generated benchmark | 41GB |
| Qwen/Qwen3-Next-80B-A3B-Instruct-FP8 | Q4 | 15.12 tok/sEstimated Auto-generated benchmark | 39GB |
| openai/gpt-oss-120b | Q4 | 15.04 tok/sEstimated Auto-generated benchmark | 59GB |
| NousResearch/Hermes-3-Llama-3.1-70B | Q4 | 14.97 tok/sEstimated Auto-generated benchmark | 34GB |
| Qwen/Qwen3-Next-80B-A3B-Thinking-FP8 | Q4 | 14.95 tok/sEstimated Auto-generated benchmark | 39GB |
| Qwen/Qwen3-30B-A3B | FP16 | 14.95 tok/sEstimated Auto-generated benchmark | 61GB |
| unsloth/gpt-oss-20b-unsloth-bnb-4bit | FP16 | 14.85 tok/sEstimated Auto-generated benchmark | 41GB |
| meta-llama/Llama-3.3-70B-Instruct | Q4 | 14.85 tok/sEstimated Auto-generated benchmark | 34GB |
| Qwen/Qwen3-30B-A3B-Instruct-2507-FP8 | FP16 | 14.82 tok/sEstimated Auto-generated benchmark | 61GB |
| lmstudio-community/Qwen3-Coder-30B-A3B-Instruct-MLX-5bit | FP16 | 14.81 tok/sEstimated Auto-generated benchmark | 61GB |
| Qwen/Qwen2.5-Math-72B-Instruct | Q4 | 14.59 tok/sEstimated Auto-generated benchmark | 36GB |
| mistralai/Mixtral-8x22B-Instruct-v0.1 | Q8 | 14.10 tok/sEstimated Auto-generated benchmark | 138GB |
| Qwen/Qwen2.5-72B-Instruct | Q4 | 14.08 tok/sEstimated Auto-generated benchmark | 36GB |
| deepseek-ai/DeepSeek-Coder-V2-Instruct-0724 | Q4 | 11.94 tok/sEstimated Auto-generated benchmark | 115GB |
| deepseek-ai/DeepSeek-Math-V2 | Q4 | 11.87 tok/sEstimated Auto-generated benchmark | 383GB |
| Qwen/Qwen3-Next-80B-A3B-Thinking | Q8 | 11.57 tok/sEstimated Auto-generated benchmark | 78GB |
| Qwen/Qwen2.5-72B-Instruct | Q8 | 11.34 tok/sEstimated Auto-generated benchmark | 71GB |
| Qwen/Qwen2.5-72B-Instruct | Q8 | 11.31 tok/sEstimated Auto-generated benchmark | 70GB |
| codellama/CodeLlama-34b-hf | FP16 | 11.21 tok/sEstimated Auto-generated benchmark | 70GB |
| meta-llama/Meta-Llama-3-70B-Instruct | FP16 | 11.05 tok/sEstimated Auto-generated benchmark | 137GB |
| Qwen/Qwen3-Next-80B-A3B-Instruct | Q8 | 11.04 tok/sEstimated Auto-generated benchmark | 78GB |
| moonshotai/Kimi-Linear-48B-A3B-Instruct | FP16 | 11.02 tok/sEstimated Auto-generated benchmark | 101GB |
| Qwen/Qwen2.5-Coder-32B-Instruct | FP16 | 10.90 tok/sEstimated Auto-generated benchmark | 67GB |
| meta-llama/Llama-3.1-70B-Instruct | Q8 | 10.81 tok/sEstimated Auto-generated benchmark | 69GB |
| Qwen/Qwen2.5-32B | FP16 | 10.66 tok/sEstimated Auto-generated benchmark | 66GB |
| NousResearch/Hermes-3-Llama-3.1-70B | Q8 | 10.63 tok/sEstimated Auto-generated benchmark | 69GB |
| moonshotai/Kimi-K2-Thinking | FP16 | 10.49 tok/sEstimated Auto-generated benchmark | 1956GB |
| Qwen/Qwen3-32B | FP16 | 10.45 tok/sEstimated Auto-generated benchmark | 66GB |
| openai/gpt-oss-120b | Q8 | 10.44 tok/sEstimated Auto-generated benchmark | 117GB |
| meta-llama/Llama-3.1-70B-Instruct | FP16 | 10.43 tok/sEstimated Auto-generated benchmark | 137GB |
| 01-ai/Yi-1.5-34B-Chat | FP16 | 10.40 tok/sEstimated Auto-generated benchmark | 70GB |
| dphn/dolphin-2.9.1-yi-1.5-34b | FP16 | 10.34 tok/sEstimated Auto-generated benchmark | 70GB |
| deepseek-ai/DeepSeek-V2.5 | FP16 | 10.26 tok/sEstimated Auto-generated benchmark | 1312GB |
| Qwen/Qwen3-Next-80B-A3B-Instruct-FP8 | Q8 | 10.21 tok/sEstimated Auto-generated benchmark | 78GB |
| Qwen/Qwen3-Next-80B-A3B-Thinking-FP8 | Q8 | 10.18 tok/sEstimated Auto-generated benchmark | 78GB |
| AI-MO/Kimina-Prover-72B | Q8 | 10.17 tok/sEstimated Auto-generated benchmark | 70GB |
| nvidia/Llama-3.1-Nemotron-70B-Instruct-HF | Q8 | 10.16 tok/sEstimated Auto-generated benchmark | 69GB |
| RedHatAI/Llama-3.2-90B-Vision-Instruct-FP8-dynamic | Q8 | 10.15 tok/sEstimated Auto-generated benchmark | 88GB |
| baichuan-inc/Baichuan-M2-32B | FP16 | 10.13 tok/sEstimated Auto-generated benchmark | 66GB |
| RedHatAI/Meta-Llama-3.1-70B-Instruct-quantized.w4a16 | FP16 | 10.07 tok/sEstimated Auto-generated benchmark | 137GB |
| meta-llama/Llama-3.3-70B-Instruct | FP16 | 10.07 tok/sEstimated Auto-generated benchmark | 137GB |
| meta-llama/Llama-3.3-70B-Instruct | Q8 | 10.03 tok/sEstimated Auto-generated benchmark | 69GB |
| MiniMaxAI/MiniMax-VL-01 | Q4 | 9.97 tok/sEstimated Auto-generated benchmark | 256GB |
| mistralai/Mistral-Large-Instruct-2411 | Q8 | 9.90 tok/sEstimated Auto-generated benchmark | 120GB |
| unsloth/DeepSeek-R1-Distill-Qwen-32B-bnb-4bit | FP16 | 9.86 tok/sEstimated Auto-generated benchmark | 66GB |
| Qwen/Qwen2.5-32B-Instruct | FP16 | 9.86 tok/sEstimated Auto-generated benchmark | 66GB |
| Qwen/Qwen2.5-Math-72B-Instruct | Q8 | 9.70 tok/sEstimated Auto-generated benchmark | 71GB |
| deepseek-ai/DeepSeek-R1-Distill-Qwen-32B | FP16 | 9.67 tok/sEstimated Auto-generated benchmark | 66GB |
| Qwen/Qwen2.5-32B-Instruct | FP16 | 9.59 tok/sEstimated Auto-generated benchmark | 67GB |
| RedHatAI/Llama-3.3-70B-Instruct-FP8-dynamic | FP16 | 9.41 tok/sEstimated Auto-generated benchmark | 137GB |
| Qwen/QwQ-32B-Preview | FP16 | 9.39 tok/sEstimated Auto-generated benchmark | 67GB |
| deepseek-ai/deepseek-coder-33b-instruct | FP16 | 9.24 tok/sEstimated Auto-generated benchmark | 68GB |
| MiniMaxAI/MiniMax-M1-40k | Q4 | 9.17 tok/sEstimated Auto-generated benchmark | 255GB |
| Qwen/Qwen3-235B-A22B | Q4 | 9.16 tok/sEstimated Auto-generated benchmark | 115GB |
| deepseek-ai/DeepSeek-Math-V2 | Q8 | 8.71 tok/sEstimated Auto-generated benchmark | 766GB |
| mistralai/Mistral-Large-3-675B-Instruct-2512 | Q4 | 8.58 tok/sEstimated Auto-generated benchmark | 378GB |
| deepseek-ai/DeepSeek-Coder-V2-Instruct-0724 | Q8 | 8.36 tok/sEstimated Auto-generated benchmark | 231GB |
| mistralai/Mixtral-8x22B-Instruct-v0.1 | FP16 | 8.00 tok/sEstimated Auto-generated benchmark | 275GB |
| mistralai/Mistral-Large-3-675B-Instruct-2512 | Q8 | 7.04 tok/sEstimated Auto-generated benchmark | 755GB |
| MiniMaxAI/MiniMax-M1-40k | Q8 | 6.88 tok/sEstimated Auto-generated benchmark | 510GB |
| MiniMaxAI/MiniMax-VL-01 | Q8 | 6.85 tok/sEstimated Auto-generated benchmark | 511GB |
| RedHatAI/Llama-3.2-90B-Vision-Instruct-FP8-dynamic | FP16 | 6.42 tok/sEstimated Auto-generated benchmark | 176GB |
| meta-llama/Llama-3.3-70B-Instruct | FP16 | 6.36 tok/sEstimated Auto-generated benchmark | 138GB |
| Qwen/Qwen3-235B-A22B | Q8 | 6.33 tok/sEstimated Auto-generated benchmark | 230GB |
| Qwen/Qwen2.5-Math-72B-Instruct | FP16 | 6.31 tok/sEstimated Auto-generated benchmark | 142GB |
| Qwen/Qwen3-Next-80B-A3B-Thinking | FP16 | 6.28 tok/sEstimated Auto-generated benchmark | 156GB |
| Qwen/Qwen2.5-72B-Instruct | FP16 | 6.23 tok/sEstimated Auto-generated benchmark | 141GB |
| Qwen/Qwen2.5-72B-Instruct | FP16 | 6.23 tok/sEstimated Auto-generated benchmark | 142GB |
| Qwen/Qwen3-Next-80B-A3B-Instruct | FP16 | 6.20 tok/sEstimated Auto-generated benchmark | 156GB |
| Qwen/Qwen3-Next-80B-A3B-Thinking-FP8 | FP16 | 6.15 tok/sEstimated Auto-generated benchmark | 156GB |
| Qwen/Qwen3-Next-80B-A3B-Instruct-FP8 | FP16 | 6.14 tok/sEstimated Auto-generated benchmark | 156GB |
| NousResearch/Hermes-3-Llama-3.1-70B | FP16 | 5.78 tok/sEstimated Auto-generated benchmark | 138GB |
| openai/gpt-oss-120b | FP16 | 5.68 tok/sEstimated Auto-generated benchmark | 235GB |
| meta-llama/Llama-3.1-70B-Instruct | FP16 | 5.61 tok/sEstimated Auto-generated benchmark | 138GB |
| mistralai/Mistral-Large-Instruct-2411 | FP16 | 5.41 tok/sEstimated Auto-generated benchmark | 240GB |
| nvidia/Llama-3.1-Nemotron-70B-Instruct-HF | FP16 | 5.29 tok/sEstimated Auto-generated benchmark | 138GB |
| AI-MO/Kimina-Prover-72B | FP16 | 5.27 tok/sEstimated Auto-generated benchmark | 141GB |
| deepseek-ai/DeepSeek-Math-V2 | FP16 | 4.21 tok/sEstimated Auto-generated benchmark | 1532GB |
| deepseek-ai/DeepSeek-Coder-V2-Instruct-0724 | FP16 | 4.01 tok/sEstimated Auto-generated benchmark | 461GB |
| MiniMaxAI/MiniMax-VL-01 | FP16 | 3.78 tok/sEstimated Auto-generated benchmark | 1021GB |
| MiniMaxAI/MiniMax-M1-40k | FP16 | 3.64 tok/sEstimated Auto-generated benchmark | 1020GB |
| mistralai/Mistral-Large-3-675B-Instruct-2512 | FP16 | 3.56 tok/sEstimated Auto-generated benchmark | 1509GB |
| Qwen/Qwen3-235B-A22B | FP16 | 3.48 tok/sEstimated Auto-generated benchmark | 460GB |
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 | 8.58 tok/sEstimated | 378GB (have 16GB) |
| mistralai/Mistral-Large-3-675B-Instruct-2512 | Q8 | Not supported | 7.04 tok/sEstimated | 755GB (have 16GB) |
| EssentialAI/rnj-1 | FP16 | Not supported | 23.44 tok/sEstimated | 19GB (have 16GB) |
| EssentialAI/rnj-1 | Q8 | Fits comfortably | 40.13 tok/sEstimated | 10GB (have 16GB) |
| EssentialAI/rnj-1 | Q4 | Fits comfortably | 61.17 tok/sEstimated | 5GB (have 16GB) |
| mistralai/Mistral-Large-3-675B-Instruct-2512 | FP16 | Not supported | 3.56 tok/sEstimated | 1509GB (have 16GB) |
| allenai/OLMo-2-0425-1B | FP16 | Fits comfortably | 32.35 tok/sEstimated | 2GB (have 16GB) |
| Qwen/Qwen3-Next-80B-A3B-Instruct | Q4 | Not supported | 16.91 tok/sEstimated | 39GB (have 16GB) |
| Qwen/Qwen2.5-3B-Instruct | Q4 | Fits comfortably | 94.24 tok/sEstimated | 2GB (have 16GB) |
| Qwen/Qwen3-4B | Q4 | Fits comfortably | 70.64 tok/sEstimated | 2GB (have 16GB) |
| meta-llama/Llama-3.3-70B-Instruct | Q8 | Not supported | 19.28 tok/sEstimated | 68GB (have 16GB) |
| meta-llama/Llama-3.3-70B-Instruct | FP16 | Not supported | 10.07 tok/sEstimated | 137GB (have 16GB) |
| Qwen/Qwen3-Embedding-8B | Q4 | Fits comfortably | 80.26 tok/sEstimated | 4GB (have 16GB) |
| Qwen/Qwen3-14B | FP16 | Not supported | 23.03 tok/sEstimated | 29GB (have 16GB) |
| Qwen/Qwen3-1.7B | FP16 | Fits (tight) | 28.53 tok/sEstimated | 15GB (have 16GB) |
| Qwen/Qwen3-32B | FP16 | Not supported | 10.45 tok/sEstimated | 66GB (have 16GB) |
| HuggingFaceTB/SmolLM2-135M | Q8 | Fits comfortably | 52.04 tok/sEstimated | 7GB (have 16GB) |
| HuggingFaceTB/SmolLM2-135M | Q4 | Fits comfortably | 72.88 tok/sEstimated | 4GB (have 16GB) |
| HuggingFaceTB/SmolLM2-135M | FP16 | Fits (tight) | 31.54 tok/sEstimated | 15GB (have 16GB) |
| zai-org/GLM-4.6-FP8 | Q8 | Fits comfortably | 52.39 tok/sEstimated | 7GB (have 16GB) |
| zai-org/GLM-4.6-FP8 | FP16 | Fits (tight) | 29.11 tok/sEstimated | 15GB (have 16GB) |
| deepseek-ai/DeepSeek-R1-0528 | Q4 | Fits comfortably | 75.30 tok/sEstimated | 4GB (have 16GB) |
| Qwen/Qwen3-0.6B-Base | Q4 | Fits comfortably | 69.95 tok/sEstimated | 3GB (have 16GB) |
| Qwen/Qwen3-0.6B-Base | Q8 | Fits comfortably | 50.75 tok/sEstimated | 6GB (have 16GB) |
| Qwen/Qwen3-0.6B-Base | FP16 | Fits comfortably | 29.56 tok/sEstimated | 13GB (have 16GB) |
| IlyaGusev/saiga_llama3_8b | Q4 | Fits comfortably | 78.79 tok/sEstimated | 4GB (have 16GB) |
| IlyaGusev/saiga_llama3_8b | Q8 | Fits comfortably | 54.76 tok/sEstimated | 9GB (have 16GB) |
| IlyaGusev/saiga_llama3_8b | FP16 | Not supported | 26.99 tok/sEstimated | 17GB (have 16GB) |
| Qwen/Qwen2.5-Coder-1.5B | Q4 | Fits comfortably | 78.62 tok/sEstimated | 3GB (have 16GB) |
| numind/NuExtract-1.5 | FP16 | Fits (tight) | 31.08 tok/sEstimated | 15GB (have 16GB) |
| openai-community/gpt2-large | FP16 | Fits (tight) | 29.13 tok/sEstimated | 15GB (have 16GB) |
| unsloth/Llama-3.2-1B-Instruct | Q4 | Fits comfortably | 84.74 tok/sEstimated | 1GB (have 16GB) |
| Qwen/Qwen2.5-Coder-7B-Instruct | FP16 | Fits (tight) | 27.85 tok/sEstimated | 15GB (have 16GB) |
| deepseek-ai/DeepSeek-R1-Distill-Qwen-7B | Q4 | Fits comfortably | 71.83 tok/sEstimated | 4GB (have 16GB) |
| unsloth/Llama-3.2-1B-Instruct | Q8 | Fits comfortably | 66.97 tok/sEstimated | 1GB (have 16GB) |
| unsloth/Llama-3.2-1B-Instruct | FP16 | Fits comfortably | 35.83 tok/sEstimated | 2GB (have 16GB) |
| lmstudio-community/DeepSeek-R1-0528-Qwen3-8B-MLX-4bit | Q4 | Fits comfortably | 80.41 tok/sEstimated | 4GB (have 16GB) |
| Qwen/Qwen3-4B-Thinking-2507 | Q8 | Fits comfortably | 51.43 tok/sEstimated | 4GB (have 16GB) |
| HuggingFaceH4/zephyr-7b-beta | Q8 | Fits comfortably | 58.36 tok/sEstimated | 7GB (have 16GB) |
| trl-internal-testing/tiny-LlamaForCausalLM-3.2 | Q4 | Fits comfortably | 71.91 tok/sEstimated | 4GB (have 16GB) |
| trl-internal-testing/tiny-LlamaForCausalLM-3.2 | FP16 | Fits (tight) | 30.45 tok/sEstimated | 15GB (have 16GB) |
| NousResearch/Meta-Llama-3.1-8B-Instruct | Q4 | Fits comfortably | 79.47 tok/sEstimated | 4GB (have 16GB) |
| NousResearch/Meta-Llama-3.1-8B-Instruct | Q8 | Fits comfortably | 56.05 tok/sEstimated | 9GB (have 16GB) |
| NousResearch/Meta-Llama-3.1-8B-Instruct | FP16 | Not supported | 31.17 tok/sEstimated | 17GB (have 16GB) |
| apple/OpenELM-1_1B-Instruct | Q4 | Fits comfortably | 84.48 tok/sEstimated | 1GB (have 16GB) |
| AI-MO/Kimina-Prover-72B | Q8 | Not supported | 10.17 tok/sEstimated | 70GB (have 16GB) |
| microsoft/Phi-3.5-mini-instruct | Q8 | Fits comfortably | 50.81 tok/sEstimated | 4GB (have 16GB) |
| microsoft/Phi-3-medium-128k-instruct | Q4 | Fits comfortably | 59.00 tok/sEstimated | 7GB (have 16GB) |
| microsoft/Phi-3-medium-128k-instruct | Q8 | Fits comfortably | 42.06 tok/sEstimated | 14GB (have 16GB) |
| microsoft/Phi-3-medium-128k-instruct | FP16 | Not supported | 23.58 tok/sEstimated | 29GB (have 16GB) |
| NousResearch/Hermes-3-Llama-3.1-70B | Q4 | Not supported | 14.97 tok/sEstimated | 34GB (have 16GB) |
| NousResearch/Hermes-3-Llama-3.1-8B | Q8 | Fits comfortably | 39.14 tok/sEstimated | 9GB (have 16GB) |
| NousResearch/Hermes-3-Llama-3.1-8B | FP16 | Not supported | 23.14 tok/sEstimated | 17GB (have 16GB) |
| trl-internal-testing/tiny-Qwen2ForCausalLM-2.5 | FP16 | Fits (tight) | 27.12 tok/sEstimated | 15GB (have 16GB) |
| meta-llama/Llama-3.2-1B-Instruct | FP16 | Fits comfortably | 35.16 tok/sEstimated | 2GB (have 16GB) |
| google-t5/t5-3b | FP16 | Fits comfortably | 37.80 tok/sEstimated | 6GB (have 16GB) |
| rednote-hilab/dots.ocr | Q4 | Fits comfortably | 75.23 tok/sEstimated | 4GB (have 16GB) |
| Qwen/Qwen2.5-72B-Instruct | Q8 | Not supported | 11.31 tok/sEstimated | 70GB (have 16GB) |
| EleutherAI/pythia-70m-deduped | Q8 | Fits comfortably | 53.31 tok/sEstimated | 7GB (have 16GB) |
| EleutherAI/pythia-70m-deduped | FP16 | Fits (tight) | 27.26 tok/sEstimated | 15GB (have 16GB) |
| bigcode/starcoder2-3b | FP16 | Fits comfortably | 35.02 tok/sEstimated | 6GB (have 16GB) |
| unsloth/gemma-3-1b-it | Q4 | Fits comfortably | 99.03 tok/sEstimated | 1GB (have 16GB) |
| openai-community/gpt2 | FP16 | Fits (tight) | 26.69 tok/sEstimated | 15GB (have 16GB) |
| openai-community/gpt2 | Q8 | Fits comfortably | 55.51 tok/sEstimated | 7GB (have 16GB) |
| Qwen/Qwen2.5-Coder-7B-Instruct | Q8 | Fits comfortably | 49.46 tok/sEstimated | 7GB (have 16GB) |
| deepseek-ai/DeepSeek-OCR | FP16 | Fits comfortably | 35.09 tok/sEstimated | 7GB (have 16GB) |
| facebook/opt-125m | Q8 | Fits comfortably | 57.60 tok/sEstimated | 7GB (have 16GB) |
| TinyLlama/TinyLlama-1.1B-Chat-v1.0 | FP16 | Fits comfortably | 35.49 tok/sEstimated | 2GB (have 16GB) |
| trl-internal-testing/tiny-Qwen2ForCausalLM-2.5 | Q4 | Fits comfortably | 82.77 tok/sEstimated | 4GB (have 16GB) |
| trl-internal-testing/tiny-Qwen2ForCausalLM-2.5 | Q8 | Fits comfortably | 50.50 tok/sEstimated | 7GB (have 16GB) |
| Qwen/Qwen3-8B | Q4 | Fits comfortably | 82.46 tok/sEstimated | 4GB (have 16GB) |
| distilbert/distilgpt2 | Q8 | Fits comfortably | 52.42 tok/sEstimated | 7GB (have 16GB) |
| distilbert/distilgpt2 | FP16 | Fits (tight) | 30.48 tok/sEstimated | 15GB (have 16GB) |
| meta-llama/Llama-3.2-3B-Instruct | FP16 | Fits comfortably | 32.26 tok/sEstimated | 6GB (have 16GB) |
| vikhyatk/moondream2 | Q4 | Fits comfortably | 81.57 tok/sEstimated | 4GB (have 16GB) |
| vikhyatk/moondream2 | Q8 | Fits comfortably | 57.86 tok/sEstimated | 7GB (have 16GB) |
| petals-team/StableBeluga2 | Q4 | Fits comfortably | 83.39 tok/sEstimated | 4GB (have 16GB) |
| petals-team/StableBeluga2 | Q8 | Fits comfortably | 56.88 tok/sEstimated | 7GB (have 16GB) |
| Qwen/Qwen3-4B | Q8 | Fits comfortably | 56.63 tok/sEstimated | 4GB (have 16GB) |
| mistralai/Mistral-7B-Instruct-v0.1 | Q4 | Fits comfortably | 76.87 tok/sEstimated | 4GB (have 16GB) |
| mistralai/Mistral-7B-Instruct-v0.1 | Q8 | Fits comfortably | 51.63 tok/sEstimated | 7GB (have 16GB) |
| mistralai/Mistral-7B-Instruct-v0.1 | FP16 | Fits (tight) | 26.50 tok/sEstimated | 15GB (have 16GB) |
| zai-org/GLM-4.5-Air | Q4 | Fits comfortably | 69.69 tok/sEstimated | 4GB (have 16GB) |
| zai-org/GLM-4.5-Air | Q8 | Fits comfortably | 51.68 tok/sEstimated | 7GB (have 16GB) |
| zai-org/GLM-4.5-Air | FP16 | Fits (tight) | 26.63 tok/sEstimated | 15GB (have 16GB) |
| unsloth/gpt-oss-20b-BF16 | Q8 | Not supported | 28.49 tok/sEstimated | 20GB (have 16GB) |
| Qwen/Qwen3-30B-A3B | FP16 | Not supported | 14.95 tok/sEstimated | 61GB (have 16GB) |
| Qwen/Qwen2.5-Coder-1.5B | Q8 | Fits comfortably | 56.22 tok/sEstimated | 5GB (have 16GB) |
| Qwen/Qwen2.5-Coder-1.5B | FP16 | Fits comfortably | 27.05 tok/sEstimated | 11GB (have 16GB) |
| rinna/japanese-gpt-neox-small | Q4 | Fits comfortably | 78.51 tok/sEstimated | 4GB (have 16GB) |
| swiss-ai/Apertus-8B-Instruct-2509 | Q4 | Fits comfortably | 83.33 tok/sEstimated | 4GB (have 16GB) |
| swiss-ai/Apertus-8B-Instruct-2509 | Q8 | Fits comfortably | 51.84 tok/sEstimated | 9GB (have 16GB) |
| swiss-ai/Apertus-8B-Instruct-2509 | FP16 | Not supported | 28.19 tok/sEstimated | 17GB (have 16GB) |
| meta-llama/Llama-2-7b-chat-hf | FP16 | Fits (tight) | 26.65 tok/sEstimated | 15GB (have 16GB) |
| Qwen/Qwen2-7B-Instruct | Q4 | Fits comfortably | 70.85 tok/sEstimated | 4GB (have 16GB) |
| Qwen/Qwen2-7B-Instruct | Q8 | Fits comfortably | 55.00 tok/sEstimated | 7GB (have 16GB) |
| ibm-granite/granite-docling-258M | Q8 | Fits comfortably | 50.81 tok/sEstimated | 7GB (have 16GB) |
| Qwen/Qwen2.5-3B | FP16 | Fits comfortably | 32.42 tok/sEstimated | 6GB (have 16GB) |
| nvidia/NVIDIA-Nemotron-Nano-9B-v2 | Q4 | Fits comfortably | 59.26 tok/sEstimated | 5GB (have 16GB) |
| AI-MO/Kimina-Prover-72B | Q4 | Not supported | 15.74 tok/sEstimated | 35GB (have 16GB) |
Note: Performance estimates are calculated. Real results may vary. Methodology · Submit real data
Explore how NVIDIA A5000 stacks up for local inference workloads.
Explore how NVIDIA A6000 stacks up for local inference workloads.
Explore how RTX 4080 stacks up for local inference workloads.
Explore how RTX 4070 stacks up for local inference workloads.
Explore how RTX 3060 12GB stacks up for local inference workloads.