Mistral AI

Mistral AI is a large language model platform from Paris: the Vibe work and coding agent, the Studio developer API, Forge fine-tuning and open-weight models.

Mistral AI lets an organisation put frontier language models to work while keeping full ownership of its data. Mistral AI SAS, founded in Paris in 2023, trains the Mistral models behind the service. One account reaches a work assistant, a developer platform and a model training service. Several of the models are also released as open weights for teams that prefer to host them. Vibe is the assistant, an AI agent for long-horizon tasks that researches, drafts and works across connected tools. Vibe CLI and Vibe for IDE bring the same agent into the terminal and the editor for agentic coding. Think mode trades speed for deeper reasoning on harder questions. Studio is where developers build, deploy and govern agents, with function calling, JSON mode, prompt caching, batch processing and embeddings. Forge trains a model on enterprise data through supervised fine-tuning and preference optimization, while Document AI and the Voxtral speech models handle OCR, transcription and voice. The platform fits developers who want an API with open-weight options and engineering teams moving to agentic coding. It also fits enterprises that need private deployments, zero data retention or inference held in a chosen region. Individuals begin on a free plan and move to a monthly Pro subscription, while teams pay per user for a shared workspace. Large organisations buy Enterprise by quote, and API calls are billed by the token on top of any plan credits. Legal and finance teams send contracts, invoices, tables and handwritten pages through Document AI and get structured data back. Support teams pass call recordings to Voxtral for transcripts and build voice agents on the same models. Engineers keep Vibe running in the terminal while it writes and refactors code, and analysts ask it to work through a spreadsheet and turn the result into a dashboard canvas. Custom MCP connectors link the agent to internal systems, so its answers draw on the company's own knowledge. For an organisation that wants to own the model it depends on, Forge finishes the job by training one on its own records.