On July 24, 2024, Mistral AI released Mistral Large 2, a 123-billion-parameter model with a 128K-token context window. Mistral said the model “vastly outperforms the previous Mistral Large, and performs on par with leading models such as GPT-4o, Claude 3 Opus, and Llama 3 405B,” reporting 84.0 percent accuracy on MMLU and describing it as “a new point on the performance/cost Pareto front of open models.”
The release leaned heavily on code and tool use. Mistral said the model was trained on a large proportion of code, supports more than 80 programming languages, and is “equipped with enhanced function calling and retrieval skills” capable of parallel and sequential function calls. To address hallucination, Mistral said it fine-tuned the model “to be more cautious and discerning in its responses” and to “acknowledge when it cannot find solutions or does not have sufficient information.” It also supports dozens of natural languages including French, German, Spanish, Arabic, Hindi, Chinese, Japanese, and Korean.
Mistral Large 2 mattered as a sign that a relatively small European lab could keep pace with the largest US labs at a fraction of the parameter count, and that strong open-weight models with serious function-calling support were viable foundations for agentic business applications.