Mistral releases Large 2 with 123B parameters for cost-efficient single-node inference

Mistral has released Mistral Large 2, a new large language model featuring 123 billion parameters designed for high-throughput single-node inference. The model supports a 128k context window and dozens of languages while offering enhanced performance in coding, reasoning, and instruction following.

The pretrained version achieves 84.0% accuracy on MMLU, setting a new point on the performance/cost Pareto front for open models.
It performs on par with leading models such as GPT-4o, Claude 3 Opus, and Llama 3 405B in code and reasoning tasks.
The model is trained to minimize hallucinations and acknowledge when it lacks sufficient information for confident answers.
Weights are available under the Mistral Research License for research and non-commercial use, with commercial self-deployment requiring a separate license.

The release aims to provide a cost-efficient alternative for building innovative AI applications, with availability on la Plateforme, HuggingFace, and major cloud providers including Google Cloud Platform, Azure, Amazon Bedrock, and IBM watsonx.ai.

Benchmarks

Benchmark	Model	Score
MMLU	Mistral Large 2	84%

Benchmark

Model

Score

MMLU

Mistral Large 2

84%

Benchmarks

Mistral launches regional endpoints, open model support, and European compute coalition

mistral.rs v0.8.10 adds /v1/skills support for local models

Mistral releases Mistral Large 2 with 128K context and improved multilingual support

Mistral AI releases Shieldstral 1.0 3B, a policy-adaptive multimodal safety classifier

Mistral releases Shieldstral-1.0-3B safety model