The Archive

Empirical study of 40+ transformer compression experiments on GPT-2 and Mistral 7B reveals variance-importance decoupling.

Mistral Studio adds Model Context Protocol support with custom connectors and approval workflows for enterprise data integration.

Mistral releases Spaces, a CLI tool designed for both human developers and autonomous agents.

Mistral AI shares design philosophy for CLI tools supporting both human users and AI agents, emphasizing unified tooling that improves developer experience.

Mistral is one of the world's leading frontier model labs, and has just launched Voxtral TTS, their latest step in their strategy to offer open frontier intelligence for every modality.

Mistral open-sources Voxtral, a fast, adaptable TTS model for voice agents with real-time synthesis.

Mistral introduces Forge, enabling enterprises to build custom frontier models fine-tuned on proprietary data.

Mistral AI joins NVIDIA Nemotron Coalition as founding member to co-develop open frontier models and multimodal capabilities.

Mistral releases Leanstral, first open-source code agent for Lean 4 formal verification.

Mistral releases OCR 3, advancing accuracy and efficiency benchmarks for document processing and layout analysis.

Mistral ships Devstral 2 and Mistral Vibe CLI, open-source agentic coding models with autonomous agent scaffolding.

Mistral releases Mistral 3, a family of open-source multimodal frontier models supporting text, image, and audio.

Mistral AI Studio is a production platform for deploying and managing AI applications with enterprise integration.

Mistral AI adds memory and control features to AI systems, enabling transparent user recall management.

Voxtral: Mistral AI releases open-source frontier speech understanding models for audio processing.

Mistral AI launches AI for Citizens initiative to help governments deploy AI for public sector innovation.

Mistral Compute offers frontier AI infrastructure access for model training and inference at scale.

Devstral: Mistral AI open-source model optimized for autonomous coding agents and software development.

Mistral Medium 3 delivers state-of-the-art performance at 8X lower cost with radically simplified enterprise deployments.

Using Mistral Models for LLM as a Judge (With Structured Outputs)

SOTA. Multimodal. Multilingual. Apache 2.0

Introducing the world’s best document understanding API.

Transform meeting transcripts into development tasks with Mistral AI: From call transcript to PRD to engineering tickets.

One of the many custom-trained models to serve specific geographies, markets, and customers

Mistral Small 3: Apache 2.0, 81% MMLU, 150 tokens/s

Mistral AI releases Codestral 25.01, a code-generation model with API access for developers.

Mistral launches free multi-capability chat interface with search, vision, ideation, and coding features.

Mistral AI announces Pixtral Large, an upgraded multimodal model for vision and text tasks.

Mistral introduces Batch API enabling lower-cost asynchronous inference for builders.

Mistral AI releases moderation API for detecting undesirable text across policy dimensions.