Paris-based Mistral AI launched Mistral Large 4 on October 6, a 1-trillion-parameter model utilizing a mixture-of-experts architecture that activates 49 billion parameters per query. The company plans to release the model's weights by the end of October, positioning it as a leading open-weight option from Europe or the US. Priced at $1.36 per million input tokens and $4.18 per million output tokens, Large 4 offers significantly lower costs than competitors like Claude Opus 5.5 and GPT-6 Astra.
The launch follows a €3 billion Series D funding round in September led by Samsung, valuing the company above €21 billion. Benchmark results show mixed performance: Large 4 scored 54.7 on Finance Agent v2, surpassing GPT-6 Astra but trailing Claude Opus 5.5. On AutomationBench, it achieved 59.9 points, while DeepSWE 1.1 coding tests yielded a score of 62, placing it behind Kimi K3 but ahead of GLM-5.3 and DeepSeek V4 Pro. The model's nickname, 'Le Chonk,' references a community meme originating from a fictional 'fat kitten' concept popularized in June.
Mistral’s decision to open-source the weights for Large 4 reinforces its strategy of providing sovereign AI infrastructure, allowing nations and enterprises to deploy frontier models without relying on closed external providers. This approach directly addresses data privacy and regulatory concerns, particularly relevant given recent high-value contracts with state-backed entities like Saudi Arabia’s HUMAIN. By offering a cost-effective alternative with transparent benchmarks, Mistral aims to capture market share from proprietary giants whose higher token prices may limit widespread adoption in budget-sensitive sectors.
However, the competitive landscape remains challenging as benchmark disparities highlight gaps in specific capabilities compared to top-tier proprietary models. While Large 4 excels in cost efficiency, its scores on complex automation and coding tasks trail leaders like Claude Opus 5.5 and Gemini 4 Argon. The true test will be whether the developer community can effectively leverage the open weights to close these performance gaps through fine-tuning and specialized applications, thereby validating the viability of open-weight models in enterprise-critical workflows.
