Mistral AI
The EU-sovereign near-frontier stack — open weights, aggressive pricing and a real agent platform; what you trade is ecosystem depth and the absolute top end, not sovereignty.
PRICING
| Free | Free | limited messages, searches, coding sessions |
| Education plan | $5.99/mo | verified students, max 12 months |
| Pro | $14.99/mo | fair-usage limits apply |
| Team | $24.99/mo | per user, $50/mo minimum |
| Enterprise | Custom | private deployments, SAML SSO |
These are the consumer/team tiers for Mistral Vibe (the assistant formerly called Le Chat). Developer use of the API is billed per token instead — Mistral Large is $2 per 1M input and $6 per 1M output tokens, with a 50% batch discount. Open-weight models can be self-hosted free under Apache 2.0, but commercial production deployments need a Mistral licence.
checked 2026-07-25 · vendor pricing page
Element scores
Strengths
The only Western lab still shipping near-frontier open weights (Mistral Large 3, Ministral 3 family) alongside a full commercial stack: the Vibe assistant (ex-Le Chat) with agent mode and 100+ connectors, Vibe for Code, Studio for building agents, Forge for custom model training, and an Agents API with MCP, handoffs and durable workflows. EU data residency in Paris is the default, not an option — the strongest sovereignty story in AI. Pricing is aggressive across the board and revenue is real: ARR grew from ~$20M to $400M+ in a year, with ASML anchoring a €1.7B Series C at €11.7B (a €3B raise at ~€20B was reported in talks in June 2026).
Honest dings
Brand and ecosystem depth trail badly — TechCrunch notes Claude is more popular than Mistral's models even at Paris' own Station F, and third-party integrations, tutorials and agent patterns are far thinner than OpenAI's. The top-end model quality is a step behind the US frontier, the eval surface is thin, and the deprecation table shows aggressive model churn (many models retired within months). The May 2026 Le Chat → Vibe rebrand muddies naming mid-adoption.
Sources (9) — every claim traceable
Every audit lists the research it rests on — transparency and traceability are the product. Tools evolve: each audit is a snapshot of its audit date, and re-audits supersede older versions (kept below for reference).
- mistral.ai/pricing — Official plans: Free / Pro $14.99 / Team $24.99/user / Enterprise / Student $5.99; API Large $2-$6 per M tokens, 50% batch discount; self-hosting license terms; product lineup (Studio, Forge, Vibe, Compute) (accessed 2026-07-08)
- docs.mistral.ai/models/overview — Current model lineup: Medium 3.5, Small 4 (hybrid instruct+reasoning), Large 3 open-weight, Ministral 3 3B/8B/14B, Voxtral TTS, OCR 4, Devstral 2, Moderation 2, Codestral Embed; deprecation/retirement table (accessed 2026-07-08)
- mistral.ai/news/agents-api — Agents API: built-in connectors (code execution, web search via FLUX image gen, document library RAG), MCP tools, persistent memory, stateful conversations, multi-agent handoffs (accessed 2026-07-08)
- docs.mistral.ai/studio-api/workflows/building-workf… — Durable agents in Workflows: crash-surviving agent state, multi-agent handoffs, MCP stdio/SSE transports, built-in tools, event history (accessed 2026-07-08)
- techcrunch.com/2026/07/04/what-is-mistral-ai-everyt… — Independent overview 04/07/2026: ARR above $400M (02/2026) vs $20M a year earlier, $1B target, Le Chat → Vibe rebrand, Koyeb acquisition, Mistral Compute, €4B data-center plan, brand-recognition criticism (accessed 2026-07-08)
- mistral.ai/news/mistral-ai-raises-1-7-b-to-accelera… — Official: €1.7B Series C at €11.7B post-money led by ASML (09/09/2025) (accessed 2026-07-08)
- techcrunch.com/2026/06/12/mistral-is-rumored-to-be-… — Reported (Bloomberg via TechCrunch): €3B raise in talks at ~€20B valuation (06/2026) — treated as rumor, not fact (accessed 2026-07-08)
- getpanto.ai/blog/mistral-ai-statistics — Independent stats roundup: ~450K customers + 1,031 high-value (07/2025 company disclosure), Le Chat Pro/Team price corroboration, $830M debt for Paris data center (03/2026) (accessed 2026-07-08)
- hyperion-consulting.io/en/insights/mistral-ai-compl… — Independent enterprise guide (03/2026): EU-only data processing on La Plateforme by default, embeddings quality vs OpenAI, hosted fine-tuning scope, guardrails/JSON mode/batch API (accessed 2026-07-08)