French startup Mistral AI has unveiled its new flagship model, called Large 2, which is significantly more capable of code generation, mathematics and reasoning. announcedMistral also says that the Large 2 has been added with improved multi-language support and advanced function call capabilities.
The Mistral Large 2 model has 123 billion parameters, which allows it to run at high throughput on a single H100 node. The model supports French, German, Spanish, Italian, Portuguese, Arabic, Hindi, Russian, Chinese, Japanese, and Korean. In terms of coding, more than 80 coding languages are supported, including Python, Java, C, C++, JavaScript, and Bash.
Mistral Large 2 is available free of charge for research and non-commercial use only, while commercial use requires a model license.
Mistral Large 2’s performance is comparable to OpenAI GPT-4o, Anthropic Claude Opus 3, and Meta Llama 3.1 405B in coding benchmarks. It outperforms Large 2, Llama 3.1 405B, and Claude 3 Opus in Wild Bench, Arena Hard, and MT Bench benchmarks. In the popular MMLU benchmark, the new model outperforms Llama 3.1 70B and is comparable to Llama 3.1 405B.
For developers, Mistral Large 2 now comes with enhanced function call and retrieval capabilities. It now executes both parallel and sequential function calls, allowing developers to build complex business AI applications.
Mistral is strengthening its model portfolio. It offers two general purpose models, Mistral Nemo and Mistral Large, and two special models, Codestral and Embed. Mistral will discontinue the Apache models (Mistral 7B, Mistral 8x7B and 8x22B, Codestral Mamba, Mathstral) in the future.
Microsoft and Mistral already had a partnership to make Mistral models available on Azure. Now they have partnered with Google to make Mistral models available on Google Cloud.
Source: www.technopat.net
