LLM Roofline
Wave 0 Model Hubs

Open-Weight LLM Sizing & Architecture Directory (32 Models)

Every model below is verified against official Hugging Face config.json tensors and live OpenRouter pricing. Click any model for full 28-GPU hardware rankings and quantization matrices.

Model Family Params (Total / Active) Attention Q4_K_M Weights Max Ctx API $/1M (In/Out)
Qwen3 235B-A22B Qwen3 235.1B / 22B MoE GQA 135.4 GiB 40k $0.14 / $0.80
Qwen3 32B Qwen3 32.8B Dense GQA 18.9 GiB 40k $0.08 / $0.28
Qwen3 30B-A3B Qwen3 30.5B / 3.3B MoE GQA 17.6 GiB 40k $0.12 / $0.50
Qwen3 14B Qwen3 14.8B Dense GQA 8.5 GiB 40k $0.12 / $0.24
Qwen3 8B Qwen3 8.2B Dense GQA 4.7 GiB 40k $0.04 / $0.14
Qwen2.5 Coder 32B Qwen2.5 32.8B Dense GQA 18.9 GiB 32k $0.66 / $1.00
Qwen2.5 72B Instruct Qwen2.5 72.7B Dense GQA 41.9 GiB 32k $0.36 / $0.40
Qwen2.5 14B Instruct Qwen2.5 14.8B Dense GQA 8.5 GiB 32k $0.06 / $0.20
Qwen2.5 7B Instruct Qwen2.5 7.6B Dense GQA 4.4 GiB 32k $0.04 / $0.12
DeepSeek-R1 671B DeepSeek 671B / 37B MoE MLA 386.4 GiB 160k $0.50 / $2.15
DeepSeek-V3 671B DeepSeek 671B / 37B MoE MLA 386.4 GiB 160k $0.29 / $1.14
DeepSeek-R1 Distill Llama 70B DeepSeek 70.6B Dense GQA 40.7 GiB 128k $0.20 / $0.60
DeepSeek-R1 Distill Qwen 32B DeepSeek 32.8B Dense GQA 18.9 GiB 128k $0.10 / $0.30
DeepSeek-R1 Distill Qwen 14B DeepSeek 14.8B Dense GQA 8.5 GiB 128k $0.06 / $0.20
Llama 4 Maverick 400B-A17B Llama 4 402B / 17B MoE GQA 231.5 GiB 1024k $0.35 / $1.00
Llama 4 Scout 109B-A17B Llama 4 109B / 17B MoE GQA 62.8 GiB 1024k $0.10 / $0.30
Llama 3.3 70B Instruct Llama 3 70.6B Dense GQA 40.7 GiB 128k $0.71 / $0.71
Llama 3.1 405B Instruct Llama 3 405.9B Dense GQA 233.8 GiB 128k $0.80 / $1.20
Llama 3.1 70B Instruct Llama 3 70.6B Dense GQA 40.7 GiB 128k $0.72 / $0.72
Llama 3.1 8B Instruct Llama 3 8.03B Dense GQA 4.6 GiB 128k $0.02 / $0.04
Gemma 3 27B IT Gemma 3 27.4B Dense GQA + SW 15.8 GiB 128k $0.08 / $0.35
Gemma 3 12B IT Gemma 3 12.2B Dense GQA + SW 7.0 GiB 128k $0.05 / $0.15
Gemma 3 4B IT Gemma 3 4.3B Dense GQA + SW 2.5 GiB 128k $0.05 / $0.10
Mistral Small 3.1 24B Mistral 24B Dense GQA 13.8 GiB 128k $0.35 / $0.56
Codestral 22B Mistral 22.2B Dense GQA 12.8 GiB 32k $0.10 / $0.30
Mixtral 8x22B Instruct Mistral 141B / 39B MoE GQA 81.2 GiB 64k $2.00 / $6.00
Mixtral 8x7B Instruct Mistral 46.7B / 12.9B MoE GQA 26.9 GiB 32k $0.24 / $0.24
Phi-4 14B Phi 14.7B Dense GQA 8.5 GiB 16k $0.07 / $0.14
GLM-4 32B 0414 GLM-4 32.3B Dense GQA 18.6 GiB 32k $0.10 / $0.30
Command A 111B Cohere 111B Dense GQA + SW 63.9 GiB 256k $2.50 / $10.00
QwQ 32B Reasoning Qwen2.5 32.8B Dense GQA 18.9 GiB 40k $0.10 / $0.30
Llama-3.3 Nemotron Super 49B Nemotron 49.9B Dense GQA 28.7 GiB 128k $0.15 / $0.40