Mistral opened a public API preview of Large 4, an open-weight multimodal model, and says it plans to publish the weights later this month.
Mistral AI announced a public preview of Mistral Large 4 on October 6, 2026, making the model available through its API while preparing to release its weights later this month. The company describes it as a multimodal mixture-of-experts model with 1.05 trillion total parameters, 52 billion active parameters and a 1.6-billion-parameter vision encoder.
Mistral says the model was trained from scratch on 3,800 NVIDIA Grace Blackwell GPUs in its European data centers. The company says it supports a one-million-token context window and was trained on multilingual data spanning more than 160 languages, including all official EU languages. Its release announcement highlights coding, tool-using agents, image understanding, cybersecurity and professional finance and legal tasks.
The company has published its own benchmark results, including a 63% score on DeepSWE 1.1, and reports strong results on cybersecurity and visual-grounding evaluations. Those claims should be read as company-reported findings: independent comparisons may differ, and Mistral says it plans to provide additional benchmarks and technical details alongside the weights. Until then, users can access the preview through Mistral’s API; the company says it is also conducting security testing with external partners before the planned weight release.
The launch gives organizations an early opportunity to evaluate a large model from a European provider, including its performance on coding and multimodal workflows. If the weights are released as planned, customers will be able to inspect and deploy them beyond the hosted preview, subject to the applicable terms. Mistral also says Large 4 will underpin future specialized models. The announcement follows the company’s recent investment in European compute infrastructure, positioning the model as both a product preview and a demonstration of its effort to build models and run them in-region.