Platform Profile
Mistral
Mistral AI builds efficient, high-performance open and closed models with a European base.
Role
Efficient frontier models
Stack
Mixture of experts, open weights
Status
Active
Mistral emphasizes performance-per-compute and open innovation.
Related Coverage
- Mistral Studio Adds Managed Agents and Apps, With the Agent Runtime Still in Public Preview
Mistral says it is expanding Studio with managed agents — including skills, sandboxes, subagents, hooks, approvals and tracing — and with enterprise apps that provide hosting, authentication and access controls, plus workflow upgrades such as event triggers, a visual graph and resumable runs. The pitch is paired with Mistral Large 4, also called Le Chonk. Documentation for Managed Agents states that the feature and the Agents SDK are in public preview. Mistral says Studio can run in the cloud, a customer's VPC or on-premises.
- Mistral Opens a Responses API in Public Preview That Uses OpenAI's Format and Keeps No Conversation State
Mistral released a Responses API in public preview on October 28 for building agents on its hosted chat models, including Mistral Large 4. Its documentation says the API uses the OpenAI Responses format and returns the model's answer, reasoning and function calls as separate output items. It is stateless: the developer's application keeps the conversation history and resends it with each request. Function calling in the preview covers functions the client runs itself.
- Mistral Launches Voxtral 3 Speech Models and a Voice-Agent API From $0.04 a Minute
Mistral released Voxtral 3 on Oct. 20: speech-to-text models covering 50+ languages, Voxtral Speech 3 (Mini TTS 3) for 35+ languages and 50+ dialects, and a LiveKit-based voice-agent SDK. Listed prices include $0.0035/min for Mini Transcribe batch and $0.04/min for the agent API with Mistral Small 4. Benchmark wins are Mistral's own.
- CERN and Mistral Sign Statement of Intent to Build What CERN Calls a 'Large Physics Model'
CERN and Mistral signed a statement of intent on Oct. 15 for a long-term partnership on foundation models for science. Its first aim is a single pretrained 'Large Physics Model', which CERN calls the first foundation model for collider physics, trained on billions of simulated LHC collisions. No budget or timeline was given.
- Update: Mistral Releases Large 4 Weights Under a Modified MIT Licence; Artificial Analysis Scores the Release One Point Above the Preview
Updated Oct. 29: Mistral released the weights of an improved Mistral Large 4 under a modified MIT licence that requires model-as-a-service providers to buy a separate commercial licence, and listed API prices of $0.70/$2.10 per million input/output tokens. Mistral reports big coding gains after three more weeks of reinforcement learning; on Artificial Analysis's new Intelligence Index v5 the release scores 26 against 25 for the preview, behind DeepSeek V4.1 Flash and GLM-5.3 at 32. Originally reported Oct. 6: Mistral opened a paid preview of the roughly 1T-parameter model and promised weights by month-end.
- Mistral Raises €3B to Keep Open-Weight Training in Europe
Mistral announced a €3 billion Series D at a post-money valuation above €21 billion. Samsung led the round, with Scaleup Europe Fund and PSG Equity as co-leads, to fund research, compute and commercial expansion.