| 両方とも前のリビジョン前のリビジョン次のリビジョン | 前のリビジョン |
| ai:localllm [2026/03/20 23:04] – [Models] oga | ai:localllm [2026/03/27 12:11] (現在) – [Models] oga |
|---|
| |
| |
| ^ ^ Model ^ parameters ^ active ^ image ^ release ^ Win11 9700X \\ DDR5 128GB \\ RTX 5060Ti 16GB ^ Linux i7-13700 \\ DDR5 96GB \\ RTX 4060Ti 16GB ^ Win11 5700X \\ DDR4 96GB \\ RX 9060 XT 16GB ^ Win11 EVO-X2 \\ 128GB \\ AI Max+ 395 ^ | ^ ^ Model ^ parameters ^ active ^ image ^ release ^ Win11 9700X \\ DDR5 128GB \\ RTX 5060Ti 16GB ^ Linux i7-13700 \\ DDR5 96GB \\ RTX 4060Ti 16GB ^ Linux 3950X \\ DDR4 128GB \\ RTX 4060Ti 16GB ^ Win11 5700X \\ DDR4 96GB \\ RX 9060 XT 16GB ^ Win11 EVO-X2 \\ 128GB \\ AI Max+ 395 ^ |
| | Alibaba | Qwen3-2507 | 235B | A22B | | 2025/07/22 | 6.8 tps (Q3_K_XL) | | | | | | Alibaba | Qwen3-2507 | 235B | A22B | | 2025/07/22 | 6.8 tps (Q3_K_XL) | | | | | |
| | OpenAI | gpt-oss-120b | 120B | A5.1B | | 2025/08/05 | 23.8 tps (MXFP4) | 24.5 tps (MXFP4) | 15.5 tps (MXFP4) | 47.5 tps (MXFP4) | | | OpenAI | gpt-oss-120b | 120B | A5.1B | | 2025/08/05 | 24.3 tps (MXFP4) | 27.6 tps (MXFP4) | 21.3 tps (MXFP4) | 16.9 tps (MXFP4) | 47.5 tps (MXFP4) | |
| | DeepSeek | DeepSeek-V3.2 | 671B | A37B | | 2025/12/01 | | | | | | | DeepSeek | DeepSeek-V3.2 | 671B | A37B | | 2025/12/01 | | | | | | |
| | Mistral AI | Mistral-large-3 | 675B | A41B | | 2025/12/02 | | | | | | | Mistral AI | Mistral-large-3 | 675B | A41B | | 2025/12/02 | | | | | | |
| | Xiaomi | MiMo V2-flash | 309B | A15B | | 2025/12/16 | | | | | | | Xiaomi | MiMo V2-flash | 309B | A15B | | 2025/12/16 | | | | | | |
| | Moonshot AI | Kimi K2.5 | 1.04T | A32B | Y | 2026/01/27 | | | | | | | Moonshot AI | Kimi K2.5 | 1.04T | A32B | Y | 2026/01/27 | | | | | | |
| | Z.ai | GLM-5 | 744B | A40B | | 2026/02/12 | | | | | | | Z.ai | GLM-5 | 744B | A40B | | 2026/02/12 | | | | | | |
| | MiniMax | MiniMax M2.5 | 230B | A10B | | 2026/02/13 | 11.6 tps (Q3_K_XL) | | | | | | MiniMax | MiniMax M2.5 | 230B | A10B | | 2026/02/13 | 11.6 tps (Q3_K_XL) | | | | | |
| | Alibaba | Qwen3.5 397b | 397B | A17B | Y | 2026/02/16 | | | | | | | Alibaba | Qwen3.5 397b | 397B | A17B | Y | 2026/02/16 | | | | | | |
| | Alibaba | Qwen3.5 122b | 122B | A10B | Y | 2026/02/24 | 16.7 tps (Q4_K_M) | 18.4 tps (Q4_K_M) | 11.7 tps (Q4_K_M) | 27.6 tps (Q4_K_M) | | | Alibaba | Qwen3.5 122b | 122B | A10B | Y | 2026/02/24 | 17.1 tps (Q4_K_M) | 18.8 tps (Q4_K_M) | 14.7 tps (Q4_K_M) | 11.9 tps (Q4_K_M) | 27.6 tps (Q4_K_M) | |
| | NVIDIA | Nemotron 3 super | 120B | A12B | | 2026/03/11 | 11.4 tps (Q4_K_M) | 12.4 tps (Q4_K_M) | 7.0 tps (Q4_K_M) | 15.7 tps (Q4_K_M) | | | NVIDIA | Nemotron 3 super | 120B | A12B | | 2026/03/11 | 11.4 tps (Q4_K_M) | 13.1 tps (Q4_K_M) | 9.7 tps (Q4_K_M) | 7.0 tps (Q4_K_M) | 15.7 tps (Q4_K_M) | |
| | Mistral | Mistral Small 4 | 119B | A6B | Y | 2026/03/17 | 13.8 tps (Q4_K_M) | 15.4 tps (Q4_K_M) | 11.3 tps (Q4_K_M) | 33.8 tps (Q4_K_M) | | | Mistral | Mistral Small 4 | 119B | A6B | Y | 2026/03/17 | 17.9 tps (Q4_K_M) | 19.2 tps (Q4_K_M) | 14.5 tps (Q4_K_M) | 14.1 tps (Q4_K_M) | 33.8 tps (Q4_K_M) | |
| |
| |
| * ctx-size 4096 | * ctx-size 4096 |
| * vision 対応でも計測時は mmproj 無し | * vision 対応でも計測時は mmproj 無し |
| | * Linux は GUI 無し |
| | |
| | ++++Link| |
| | |
| | * [[https://huggingface.co/Qwen]] |
| | * [[https://huggingface.co/openai]] |
| | * [[https://huggingface.co/deepseek-ai]] |
| | * [[https://huggingface.co/mistralai]] |
| | * [[https://huggingface.co/XiaomiMiMo]] |
| | * [[https://huggingface.co/moonshotai]] |
| | * [[https://huggingface.co/zai-org]] |
| | * [[https://huggingface.co/MiniMaxAI]] |
| | * [[https://huggingface.co/nvidia]] |
| | |
| | |
| | ++++ |
| | |
| |
| |
| |
