All models
33 canonical models, grouped by family. Click any name to see every backend × hardware combination we have benchmark data for. See also all hardware.
DeepSeek
- DeepSeek-Coder-V2-Lite-Instruct · 16B-A2.4B · moe-coder
- DeepSeek-V3 · 671B-A37B · moe
- DeepSeek-V3.2-Exp · 671B-A37B · moe
- Gemma-2-27b-it · 27B · instruct
- Gemma-2-9b-it · 9B · instruct
IBM Granite
- Granite-3.2-Code-8B · 8B · coder
Meta
- Llama-3.1-405B-Instruct · 405B · instruct
- Llama-3.1-70B-Instruct · 70B · instruct
- Llama-3.1-8B-Instruct · 8B · instruct
- Llama-3.2-3B-Instruct · 3B · instruct
- Llama-3.3-70B-Instruct · 70B · instruct
- Llama-4-Scout · 109B-A17B · moe
Microsoft
- Phi-4 · 14B · instruct
Mistral
- Codestral-25B · 22B · coder
- Mistral-Small-3.1-24B-Instruct · 24B · instruct
- Mixtral-8x7B-Instruct · 47B-A12.9B · moe
Moonshot
- Kimi-K2 · 1T-A32B · moe
OpenAI
- gpt-oss-120b · 120B · instruct
- gpt-oss-20b · 20B · instruct
Qwen
- Qwen2.5-0.5B-Instruct · 0.5B · instruct
- Qwen2.5-1.5B-Instruct · 1.5B · instruct
- Qwen2.5-14B-Instruct · 14B · instruct
- Qwen2.5-32B-Instruct · 32B · instruct
- Qwen2.5-3B-Instruct · 3B · instruct
- Qwen2.5-72B-Instruct · 72B · instruct
- Qwen2.5-7B-Instruct · 7B · instruct
- Qwen2.5-Coder-14B-Instruct · 14B · coder
- Qwen2.5-Coder-32B-Instruct · 32B · coder
- Qwen2.5-Coder-7B-Instruct · 7B · coder
- Qwen3-Coder-30B-A3B-Instruct · 30B-A3B · moe-coder
- Qwen3-Coder-Next · 80B-A3B · moe-coder
- Qwen3-Next-80B-A3B-Instruct · 80B-A3B · moe
Zhipu
- GLM-4.6 · 355B-A32B · moe-coder