Mistral Large 4: What a 1T Open-Weight Model Means for You

Mistral AI released Mistral Large 4 on October 6, 2026, and it is the company's biggest model by a wide margin. Mistral Large 4 has 1 trillion total parameters, natively understands both text and images, and is headed for an open-weight release by the end of this month. For developers who want frontier-scale capability they can eventually host themselves, this is one of the most significant open-weight launches of the year.

• Size: 1 trillion total parameters with 49 billion active per token, in a mixture-of-experts design that combines instruct and reasoning modes in one model.
• Price: $1.36 per million input tokens and $4.18 per million output tokens during the API public preview.
• Languages: more than 160, including every official language of the European Union.
• Open weights: Mistral says the weights will be released by the end of October 2026.

What Is Mistral Large 4?

Mistral Large 4 is a hybrid instruct-and-reasoning mixture-of-experts model. Only about 49 billion of its 1 trillion parameters fire for any given token, which is how a model this large stays practical to serve. It is natively multimodal, handling text and vision together rather than bolting an image encoder on afterward. Mistral says it trained the model on 3,800 NVIDIA Grace Blackwell GPUs in its own European datacenters. TechCrunch reports that the company nicknamed the model "Le Chonk," which is a refreshingly honest description of a trillion-parameter system.

How Does Mistral Large 4 Perform on Benchmarks?

Mistral published a broad set of results, with an emphasis on coding, agents and security work. All of these figures come from Mistral and have not yet been independently reproduced:

• DeepSWE v1.1, a software engineering test: 61.7%.
• Terminal Bench 4.0, which measures command-line agent work: 28.3%.
• Cybench: 93% of challenges solved, plus 82% on a vulnerability reproduction and patching test.
• AutomationBench: 59.9%.
• Dense 200 visual grounding: 42%, which Mistral says edges out GPT-6-Astra at 41%.

Mistral also says the model ranks in the top five globally on the Artificial Analysis cyber index. TechCrunch notes that Mistral is targeting cybersecurity, finance and chip design as early use cases, which lines up with the benchmarks it chose to highlight.

When Can You Download the Open Weights?

Right now, Mistral Large 4 is available as a public preview API through Mistral Studio. Mistral says the weights will follow by the end of October, and TechCrunch reports the company is running safety testing first. Mistral has not yet published the context window or the license terms, so teams planning to self-host should wait for the weight release before committing. Even at 49 billion active parameters, a 1T model will need serious multi-GPU hardware to run locally, so most developers will meet it through the API first.

Why This Open-Weight LLM Matters

Open-weight frontier models give companies a way to keep sensitive data inside their own infrastructure while still using top-tier capability. Mistral has been building toward this moment, from its Mistral Small 4 work inside Firefox to the funding round that bankrolled its compute push. A trillion-parameter open model from a European lab broadens the options for every developer who prefers to own their stack. For more model launches, see our AI coverage.

Sources: Mistral AI — October 6, 2026; TechCrunch — October 6, 2026.