Mistral Large 4 Release Notes

Mistral Large 4 is a state-of-the-art, open-weight, general-purpose multimodal model designed to provide a high-performance alternative to closed-source frontier models and Chinese open-weight models. It is particularly distinguished by its strong performance in cybersecurity and visual grounding tasks.

Model Architecture and Specifications

Mistral Large 4 utilizes a granular Mixture-of-Experts (MoE) architecture to balance total capacity with computational efficiency during inference.

  • Total Parameters: 1.05 Trillion
  • Active Parameters: 49 Billion
  • Vision Encoder: 1.6 Billion parameters
  • Context Window: 1 Million tokens
  • Training Infrastructure: The model was trained from scratch using 3,800 NVIDIA Grace Blackwell GPUs within Mistral's European datacenters.

Performance and Benchmarking

Mistral Large 4 demonstrates competitive capabilities across several specialized domains, though it remains slightly behind the absolute Pareto frontier in general reasoning.

Cybersecurity and Technical Proficiency

The model is positioned as a strong defender model for cybersecurity, outperforming several Chinese models in cyber benchmarks. It reports an 82% score on CyberGym-E2E. Users have noted that its lack of restrictive safety filters on cyber capabilities makes it a potent tool for both offensive and defensive security research.

Visual Grounding and Multimodality

Mistral Large 4 shows impressive visual grounding capabilities, scoring 42% on Dense 200, slightly exceeding the 41% reported for GPT-6 Astra.

Data Analytics and Programming

Early testing indicates a significant generational leap over previous Mistral iterations. In Plotly's data analytics benchmark, the model improved from 58% to 74% accuracy compared to Mistral Medium 3.5. In programming tests using the Killswitch language, it performs similarly to Kimi K3 and Claude Sonnet 5.5, although it is slower and more expensive than the latter.

Pricing and Availability

Mistral Large 4 is available in public preview (v 26.10) as of October 6, 2026.

Metric Price (per M Tokens)
Input $0.68
Cached Input $0.07
Output $2.09

While cheaper than models like Kimi K3 and GLM 5.3, it remains more expensive than "flash" variants such as GLM 5.3-flash.

Community Insights and Analysis

Industry observers and users on Hacker News have highlighted several key themes regarding this release:

  • European AI Sovereignty: Many view Mistral Large 4 as a critical step for European technological independence, providing a model trained and hosted within the EU that aligns with local regulations.
  • Hardware Efficiency: There is significant discussion regarding the efficiency of the training run, with some questioning the necessity of massive US-based superclusters given that a 1T parameter model was trained on only 3,800 Blackwell GPUs.
  • Comparison to Competitors: While praised for its cyber and vision capabilities, some users argue that Mistral's progress is slower than that of frontier labs or Chinese open-weight models like DeepSeek or GLM.

"If the vision model is truly as good as astra, that would make it best in the world. Also strong on cyber benchmarks... this is a good defender model."

"Mistral’s progress, regrettably, feels much slower. This model doesn’t knock anybody’s socks off. It’s... mediocre, and this mediocrity has also arrived months late."

Supported Features

Mistral Large 4 supports a comprehensive suite of developer tools via its API, including:

  • Structured Outputs
  • Function Calling
  • Document Q&A
  • Batching
  • Agents & Conversations

Sources

Related

  • Dispatch
    Mistral Large 4 Release Notes

    Mistral AI has launched a public preview of Mistral Large 4, a 1 trillion-parameter natively multimodal open-weight model designed for frontier performance in cybersecurity, agentic coding, and professional knowledge work.

  • Dispatch
    Mistral Large 4 Release Notes

    Mistral Large 4 (ML4), a 1 trillion-parameter multimodal model with 49 billion active parameters, is now in public preview with open weights scheduled for release by the end of October 2026.

  • Dispatch
    Pixtral Large Release Notes

    Mistral AI has released Pixtral Large, a 124B open-weights multimodal model that achieves state-of-the-art performance on MathVista, DocVQA, and VQAv2 while maintaining the text capabilities of Mistral Large 2.

  • Dispatch
    Mistral Large 2 release notes / what's new

    Mistral AI has released Mistral Large 2, a 123B parameter model designed for high throughput single-node inference with competitive performance in coding, reasoning, and multilingual capabilities.

  • Dispatch
    Mistral Medium 3 Release Notes

    Mistral AI has released Mistral Medium 3, a frontier-class model that balances state-of-the-art performance with 8x lower costs and flexible enterprise deployment options.