Mistral AI releases Large 4 weights for self-hosting
The AMW Read
The weight release delivers Mistral's announced deployment option and advances its open-weight strategy, with potential segment-wide implications for enterprise model choice.
Mistral AI releases Large 4 weights for self-hosting
Mistral AI has released downloadable weights for Large 4, its trillion-parameter multimodal model, after previewing it earlier in October. The release adds low, medium, and high reasoning modes, with high as the default. Mistral says three additional weeks of reinforcement learning improved reasoning efficiency. Developers can also access the model through Mistral's API and Studio, OpenRouter, Microsoft Foundry, and NVIDIA-hosted endpoints.
The release advances Mistral's open-weight strategy in the foundation-model market. Self-hosting and customization give enterprises another deployment option alongside hosted frontier models, though the release alone does not establish a performance or cost advantage in production. Mistral reports that Large 4 scored 48% on Artificial Analysis's Coding Agent Index while using 47% fewer tokens per task than GLM-5.3. The company also reports a 51% score on the Artificial Analysis Cyber Index and acknowledges potential uplift for real attacks observed in live-traffic monitoring.
For builders, the reasoning modes offer a way to test cost against performance on their own workloads. API pricing starts at $0.70 per million input tokens and $2.10 per million output tokens. The modified MIT license permits free use, fine-tuning, and commercialization, but model-as-a-service businesses need a separate commercial license. Teams should assess those terms alongside deployment requirements and cyber safeguards before choosing a hosting model.
