Mistral Large 4
Mistral Large 4 was trained from scratch on 3,800 NVIDIA Grace Blackwell GPUs Mistral Studio provides the preview API for Mistral Large 4 Mistral Forge offers the training, customization, and RL environment for Mistral Large 4 Artificial Analysis Cyber Index ranks Mistral Large 4 among the top five models globally Cybench ranks Mistral Large 4 with a score of 93% DeepSWE v1.1 ranks Mistral Large 4 with a score of 61.7% SWE-Atlas-QnA ranks Mistral Large 4 with a score of 59.4% Terminal-Bench 4 ranks Mistral Large 4 with a score of 28.3% Mistral Large 4 surpassed DeepSeek V4 Pro 0813 in the Coding Agent Index Mistral Large 4 surpassed Qwen3.8 Max in the Coding Agent Index Mistral Large 4 was ahead of Kimi K3 in the blind human evaluation Mistral Large 4 was ahead of GLM-5.3 in the blind human evaluation Mistral Large 4 was ahead of GLM-5.2 in the blind human evaluation Claude Opus 5 surpassed Mistral Large 4 in the blind human evaluation Mistral Large 4 surpassed GPT-6-Astra on Dense 200 visual grounding Surge AI ranked Mistral Large 4 second among five models in a blind human evaluation SciCode-Verified ranks Mistral Large 4 as state‑of‑the‑art among open‑weight models Mistral Large 4 is accelerated by NVIDIA CUDA-X libraries as part of its physics modelling pipelines. Mistral Large 4 is accelerated by NVIDIA Warp. Mistral Large 4 is also known as Le Chonk. Mistral Large 4, with approximately 1 trillion total parameters, was trained on 3,800 NVIDIA Grace Blackwell GPUs in Mistral-operated data centers in Europe.· Graphiti 生成的实体摘要
关系变化
按生效时间- Mistral Large 4ACCELERATED BYNVIDIA Warp
Mistral Large 4 is accelerated by NVIDIA Warp.
- Mistral Large 4ACCELERATED BYNVIDIA CUDA-X libraries
Mistral Large 4 is accelerated by NVIDIA CUDA-X libraries as part of its physics modelling pipelines.
- Mistral Large 4ALSO KNOWN ASLe Chonk
Mistral Large 4 is also known as Le Chonk.
- DeepSWE v1.1RANKSMistral Large 4
DeepSWE v1.1 ranks Mistral Large 4 with a score of 61.7%
- Mistral Large 4SURPASSEDGLM-5.3
Mistral Large 4 was ahead of GLM-5.3 in the blind human evaluation
- Mistral Large 4SURPASSEDGLM-5.2
Mistral Large 4 was ahead of GLM-5.2 in the blind human evaluation
- Terminal-Bench 4RANKSMistral Large 4
Terminal-Bench 4 ranks Mistral Large 4 with a score of 28.3%
- Claude Opus 5SURPASSEDMistral Large 4
Claude Opus 5 surpassed Mistral Large 4 in the blind human evaluation
- CybenchRANKSMistral Large 4
Cybench ranks Mistral Large 4 with a score of 93%
- SWE-Atlas-QnARANKSMistral Large 4
SWE-Atlas-QnA ranks Mistral Large 4 with a score of 59.4%
- Mistral Large 4SURPASSEDDeepSeek V4 Pro 0813
Mistral Large 4 surpassed DeepSeek V4 Pro 0813 in the Coding Agent Index
- Mistral Large 4SURPASSEDKimi K3
Mistral Large 4 was ahead of Kimi K3 in the blind human evaluation
- Artificial Analysis Cyber IndexRANKSMistral Large 4
Artificial Analysis Cyber Index ranks Mistral Large 4 among the top five models globally
- SciCode-VerifiedRANKSMistral Large 4
SciCode-Verified ranks Mistral Large 4 as state‑of‑the‑art among open‑weight models
- Mistral Large 4SURPASSEDQwen3.8 Max
Mistral Large 4 surpassed Qwen3.8 Max in the Coding Agent Index
- Mistral ForgePROVIDES TRAINING ENV FORMistral Large 4
Mistral Forge offers the training, customization, and RL environment for Mistral Large 4
- Mistral Large 4SURPASSEDGPT-6 Astra
Mistral Large 4 surpassed GPT-6-Astra on Dense 200 visual grounding
- Surge AIRANKSMistral Large 4
Surge AI ranked Mistral Large 4 second among five models in a blind human evaluation
- Mistral Large 4TRAINED ONNVIDIA Grace Blackwell GPUs
Mistral Large 4 was trained from scratch on 3,800 NVIDIA Grace Blackwell GPUs
- Mistral StudioPROVIDES API FORMistral Large 4
Mistral Studio provides the preview API for Mistral Large 4
- Mistral Large 4TRAINED ONNVIDIA Grace Blackwell GPUs
Mistral Large 4, with approximately 1 trillion total parameters, was trained on 3,800 NVIDIA Grace Blackwell GPUs in Mistral-operated data centers in Europe.
关系是 Graphiti 从报道中提取的事实陈述，不是原文引述；失效时间表示后来的报道显示这条关系已不成立。