Mistral Large 4 is Europe's trillion-parameter answer to US models that refuse security work
Mistral has released a public preview of Mistral Large 4 (ML4), nicknamed "le Chonk," its largest model to date with one trillion parameters. The API is available now through Mistral Studio, and model weights are expected at the end of October.
According to Mistral, ML4 is a natively multimodal model with one trillion parameters, 49 billion of them active. The company says it is the best open-weight model from the US or Europe across aggregated benchmarks, claims state-of-the-art performance in cyber defense, manufacturing and finance, and says it beats closed frontier models in visual grounding. Its context window spans one million tokens, and documentation describes a fine-grained mixture-of-experts architecture with 1.05 trillion total parameters and a 1.6 billion parameter vision encoder.
On the independent Artificial Analysis Intelligence Index, which aggregates ten benchmarks, ML4 scores 38 points. Predecessor Mistral Large 3 scored 9, and Mistral Medium 3.5 scored 14. ML4 also edges past Z.ai's GLM-5.2. Claude Opus 5.5 (Max) leads at 58 points. ML4 remains listed as proprietary because its weights have not been released.
Mistral devotes the bulk of its announcement to IT security. In the Artificial Analysis Cyber Index, the company says ML4 ranks among the top five models worldwide and leads among open-weight models developed outside China. On a test requiring a model to reproduce and patch a real open-source vulnerability, ML4 scores 82 percent, the highest of any model. According to Mistral, Claude Opus 5.5 and GPT-6 Astra score near zero because they refuse the task entirely. Mistral says ML4 refuses malicious cyber prompts more often than any other open model, citing JailbreakBench, StrongREJECT and AgentHarm, and blocks 93.3 percent of attacks in Lakera's B3 AI Security Benchmark. During the preview, Mistral charges $0.68 per million input tokens and $2.09 per million output tokens; cached inputs cost $0.07.
In coding, ML4 scores 49.8 percent on the Artificial Analysis Coding Agent Index, ahead of Deepseek V4 Pro and Qwen3.8 Max. In a blind Surge AI evaluation, annotators rated it second of five models with 3.74 of 5 points, behind Claude Opus 5 at 4.22.
Mistral says ML4 was trained from scratch on 3,800 Nvidia Grace Blackwell GPUs in its own European data centers, using about 3,000 GPUs for reinforcement learning post-training, generating roughly 33 billion tokens daily, with training data covering more than 160 languages. The RL run behind the preview is still ongoing with no signs of plateauing. Until weights ship, Mistral is red-teaming the model with security firms, vetted partners and government agencies, which receive a version with reduced moderation and expanded cyber capabilities.
Based on reporting from the original publisher. Visit the source for full context and later updates.
Publisher excerpt
Mistral's Large 4 is the company's biggest model yet, with one trillion parameters trained on its own European infrastructure. In the independent Intelligence Index, the model makes a big leap forward but still falls well short of Claude, GPT-6, and Chinese competitors. Mistral's main pitch is cybersecurity work that closed US models refuse to do. The article Mistral Large 4 is Europe's trillion-parameter answer to US models that refuse security work appeared first on The Decoder .