NeuralStackMistral AI released Mistral Large 3 on July 15 2026 with a 2 million token context window and 92 percent score on MMLU. The model achieves 89 percent on HumanEval and supports native tool calling with 28 languages. Training completed on 1.8 trillion tokens using custom H100 clusters in France.
Developers can access the model via Mistral API starting at $8 per million input tokens. Early benchmarks show 15 percent gains over the prior version in long-document summarization tasks. The release includes an open weights version under a research license for non-commercial use.
Background includes Mistral's prior focus on efficient European models after its 2023 founding. The company previously raised funds from investors including Microsoft and BNP Paribas. This launch follows regulatory pressure in the EU for transparent AI development.
Why this matters
The release intensifies competition in the high-context model segment and gives European developers a compliant alternative. Enterprises gain options for data residency while meeting upcoming EU transparency rules. Long context capabilities enable new applications in legal and scientific document analysis.
Industry observers expect Mistral to capture additional market share from US providers in regulated sectors. The open weights tier may accelerate academic research on efficient scaling laws.