Mistral releases the weights for Mistral Large 4, or Le Chonk, and says it rivals top open models such as GLM-5.3 on the Artificial Analysis Coding Agent Index

Open-source models are now matching proprietary AI performance for coding tasks, potentially lowering costs for businesses.

Mistral AI has released the weights for its latest open-source model, Mistral Large 4 (ML4), nicknamed "Le Chonk." This trillion-parameter model has demonstrated strong performance, particularly in coding tasks. On the Artificial Analysis Coding Agent Index, ML4 rivals top open models like GLM-5.3, scoring 48 out of a possible 100, just one point behind GLM-5.3's score of 49. The model also showed a significant improvement on the Terminal-Bench 4.0 coding benchmark, jumping from 28% to 41% in a short period. ML4 also boasts efficiency, using an average of 47% fewer tokens per task compared to other leading models. Additionally, the model has shown prowess in cybersecurity, outperforming DeepSeek V4.1 Flash on the CyberGym benchmark for defensive capabilities.

The release of Mistral Large 4 signifies a maturing open-source AI landscape where models are not only competitive in raw performance but also more efficient in token usage. This efficiency directly translates to reduced operational costs for developers and enterprises looking to integrate advanced AI capabilities without the premium price tag of closed-source alternatives. The improved performance on coding benchmarks also highlights a growing specialization within open models, catering to specific industry needs like software development and cybersecurity.

Mistral's emphasis on cybersecurity capabilities within ML4, coupled with its open-source nature, positions it as a significant player in the defense against AI-driven threats. As proprietary models face scrutiny for their potential security risks, the ability to deploy and customize open models like ML4 offers greater control and transparency for organizations. This move challenges the traditional divide between open and closed AI systems, suggesting a future where open models play a crucial role in both innovation and security.