Mistral AI Releases 'Le Chonk' Mistral Large 4 Open Weights with Reduced API Pricing

Mistral AI released the open weights of its flagship model, Mistral Large 4, nicknamed "Le Chonk," on October 29, 2026, alongside significant API price cuts and a strategic positioning for "sovereign AI" initiatives.

Mistral Large 4: Architecture and Capabilities

Mistral Large 4 is a natively multimodal mixture-of-experts (MoE) model, featuring approximately 1 trillion total parameters with about 52 billion active parameters per step. This architecture is designed for efficiency and performance across various tasks. The model is now available on Hugging Face, allowing broader access for research and development.

In terms of performance, Mistral Large 4 claims the top position among open-weight models for agentic coding. It ties GLM-5.3 on the Artificial Analysis Coding Agent Index while utilizing 47% fewer tokens per task. However, it trails Claude Opus 5.5 in overall rankings, with Opus 5.5 scoring higher on Cognition's FrontierCode 2.0 and the Coding Agent Index. Mistral Large 4 achieved a score of 19.7% on FrontierCode 2.0 and 41% on Terminal-Bench 4.0.

API Pricing and Licensing Details

Alongside the open-weight release, Mistral AI has significantly reduced the API prices for Mistral Large 4. The new pricing is set at $0.70 per million input tokens and $2.10 per million output tokens, a notable decrease from the preview rates of $1.36 and $4.18, respectively. This adjustment aims to encourage wider adoption and integration of the model into commercial applications. For more details on pricing, visit our AI tools pricing guide.

The model is released under a modified MIT license. This license permits free use, modification, and commercial deployment. However, offering Mistral Large 4 as a commercial cloud service requires a separate license directly from Mistral AI. This licensing structure allows Mistral AI to maintain control over its commercial cloud offerings while still fostering an open ecosystem for development.

Hardware Requirements and Sovereign AI Ambitions

Self-hosting the standard version of Mistral Large 4 requires substantial hardware, specifically eight Nvidia B300 GPUs. For those seeking a more resource-efficient option, a compressed variant is available that can run on four GB300 GPUs. These requirements highlight the computational intensity of deploying such a large model.

Mistral AI emphasizes that Mistral Large 4 was trained from scratch on European infrastructure, framing it as a cornerstone for "sovereign AI." This initiative aims to provide AI capabilities that are developed and controlled within Europe, addressing concerns about data privacy and technological independence. The model is also accessible via Microsoft Foundry, expanding its deployment options.

Cybersecurity Performance and Competitive Landscape

Mistral Large 4 demonstrates notable performance in cybersecurity, scoring 51% on the Artificial Analysis Cyber Index. This places it third among open models, behind DeepSeek V4.1 Flash (57%) and MiMo-V2.6-Pro (56%). A key highlight is its claimed 96.6% refusal rate on malicious cyber requests, which Mistral AI states is the highest among open models. This feature is critical for applications requiring robust security and ethical AI behavior.

The competitive landscape for open-weight models is dynamic. While Mistral Large 4 currently holds a strong position, Moonshot plans to release the weights for Kimi K3.1 Pro by November 4, 2026. Kimi K3.1 Pro currently leads Terminal-Bench 4.0 with 67% and the Cyber Index with 64%, suggesting that Mistral's lead in certain benchmarks may be short-lived. This ongoing competition drives rapid advancements in the field of artificial intelligence news.

Conclusion

Mistral AI's release of Mistral Large 4, "Le Chonk," marks a significant development in the open-weight AI model space. With its advanced architecture, competitive performance in agentic coding and cybersecurity, and reduced API pricing, it offers a powerful tool for developers and organizations. The strategic focus on "sovereign AI" and training on European infrastructure also underscores broader industry trends. As new models like Kimi K3.1 Pro emerge, the competition will continue to push the boundaries of what open-weight AI can achieve.

Sources

• mistralai/Mistral-Large-4-1T-A52B · Hugging Face
• Welcome Le Chonk: Mistral Large 4 on Hugging Face
• What Is Mistral Large 4? Architecture, Context, Access, and Cost
• Mistral releases weights for its trillion-parameter AI model, targeting cybersecurity | TechCrunch
• Mistral’s new 1T model aims to leapfrog closed and open rivals | TechCrunch