Mistral Large 4: Features, Access and Capabilities

What is Mistral Large 4?

Mistral Large 4 is a 1-trillion-parameter open-weight multimodal AI model created by European AI firm Mistral AI. Mistral Large 4 operates with 49 billion active parameters and is designed for enterprise applications including software engineering, cybersecurity, and document reasoning.

Mistral AI trained Mistral Large 4 from scratch on 3,800 NVIDIA Grace Blackwell GPUs located in European datacenters. The model training dataset spans over 160 languages, including every official language of the European Union. Its public preview API launched on 6 October 2026, with open weights scheduled for release by the end of October 2026.

How does Mistral Large 4 perform in coding and workflows?

Mistral Large 4 delivers strong benchmark results across software engineering, terminal execution, and automated business workflows. Mistral AI reports that Mistral Large 4 outperforms competing open-weight systems on several critical industry evaluations.

In blind human evaluations conducted by Surge AI on coding output quality, professional annotators gave Mistral Large 4 a score of 3.74 out of 5. This rating placed Mistral Large 4 ahead of Kimi K3 at 3.59, GLM-5.3 at 3.60, and GLM-5.2 at 3.40, trailing only Claude Opus 5 at 4.22.

What are the cybersecurity capabilities of Mistral Large 4?

Mistral Large 4 ranks among the top five AI models globally on the Artificial Analysis Cyber Index for cybersecurity tasks. Mistral Large 4 achieves an 82% success rate on reproducing and patching software vulnerabilities, which is the highest score reported on that evaluation.

On the Cybench benchmark of 40 cybersecurity competition challenges, Mistral Large 4 solves 93% of exercises. Internal testing by Mistral AI shows Mistral Large 4 is effective for analyzing malware, prioritizing software vulnerabilities, and writing detection rules. During its red-teaming phase, vetted partners and state authorities access Mistral Large 4 with expanded cyber capabilities to test security operations under controlled conditions.

What multimodal tasks can Mistral Large 4 handle?

Mistral Large 4 includes native vision capabilities to process documents, charts, natural scenes, and technical diagrams. Mistral Large 4 combines visual grounding with automated reasoning to inspect high-resolution imagery for specialized sectors.

The model can process gigapixel satellite imagery for emergency response scenarios and analyze complex engineering drawings by zooming in on precise components. Mistral AI designed these visual features to support industries such as manufacturing, logistics, earth observation, and pharmaceuticals.

How to access and use Mistral Large 4

Access to Mistral Large 4 is currently available through a public preview API on the Mistral Studio platform. Developers and enterprise users can test Mistral Large 4 online before its open weights are released in late October 2026.

1. Try the public preview API on Mistral Studio.
2. Access the model with reduced moderation and expanded capabilities if you are an authorized partner, state authority, or cybersecurity leader during the red-teaming phase.
3. Get the open weights once Mistral AI releases them at the end of October 2026.

Frequently asked questions

Sources: Mistral