StorySoftware
Mistral releases Mistral Large 2, a 123B-parameter model
Mistral released Mistral Large 2, a 123-billion-parameter model with 128K-token context supporting dozens of natural languages and 80+ programming languages. It performs comparably to GPT-4o, Claude 3 Opus and Llama 3 405B on coding; weights are available under a research licence with commercial self-hosting requiring a paid licence.
- Achieves 84.0% accuracy on MMLU benchmark; performs on par with GPT-4o and Claude 3 Opus on coding and reasoning.
- Trained to be cautious and discerning, minimizing hallucinations and ensuring reliable outputs.
- Available via Mistral's platform and major cloud providers: GCP Vertex AI, Azure AI Studio, Amazon Bedrock, IBM watsonx.ai.
- Commercial self-deployment requires separate Mistral Commercial License through direct contact.