Mistral AI iconMistral AI

Paris AI lab and API platform: Mistral Large 4 with 1M context at $0.70/$2.10 per 1M tokens, plus OCR, Voxtral speech, Codestral and Studio

1M Context (Mistral Large 4)
EUR 21B+ Valuation (Sept 2026)
2023 Founded

Overview

Mistral AI is a Paris company founded in April 2023 by Arthur Mensch, Guillaume Lample and Timothee Lacroix. Its API platform, managed in Mistral Studio, sells its own models by the token. The flagship is Mistral Large 4 (version 26.10): a mixture-of-experts model with about 1 trillion parameters, 49B active per token, a 1M-token context and text plus image input, priced at $0.70 input and $2.10 output per 1M tokens. Its weights are published under the Mistral Ecosystem Open Weight License, but running it as a hosted model service needs a commercial license. Mistral Small 4 and Ministral 3 cover low-cost use, and premier API-only models handle OCR (OCR 4.1), speech (Voxtral Transcribe 3 and TTS 3), code (Codestral), embeddings and moderation. The model list changes fast: Mistral Large 2, Mistral 7B and Mixtral were retired in 2025, Medium 3 and 3.1 in August 2026, and Medium 3.5 is deprecated and set to retire in April 2027. In September 2026 Mistral raised EUR 3B in a Samsung-led Series D at a post-money valuation above EUR 21B. Also from Mistral AI: Mistral open-weight models, Le Chat, its assistant, and Mistral Vibe, its coding agent.

The Verdict

Who Should Use Mistral AI?

Best For

  • European teams that want a European model vendor
  • Cost-sensitive apps: Large 4 is priced well below many frontier APIs
  • Document pipelines that need OCR priced per page
  • Teams that may later self-host the same model family

Not Ideal For

  • Teams that need the top scores on hard reasoning and coding benchmarks
  • Apps that cannot handle frequent model retirements and migrations
  • Buyers who rely on fast, hands-on vendor support

What's Great

  • Large 4 at $0.70 / $2.10 per 1M tokens with a 1M context
  • Batch jobs at 50% off; cached input up to 90% off
  • Full stack from one vendor: chat, code, OCR, speech, embeddings, moderation
  • Moderation 2 is free to call
  • G2 reviewers praise value for money and speed

Watch Out For

  • Heavy model churn: Devstral 2, Magistral 1.2 and Medium 3.x were all retired or deprecated in 2026
  • The Medium 3.5 launch drew a lukewarm reception
  • G2 reviewers report weaker results on complex tasks
  • Self-hosting Large 4 as a service needs a separate commercial license

Pricing

View all features & details

Current Models

  • Mistral Large 4 (26.10): ~1T parameters, 49B active, 1M context
  • Mistral Small 4 (26.03): Apache 2.0
  • Ministral 3 14B, 8B, 3B (25.12): $0.20, $0.15 and $0.10 per 1M
  • Codestral (25.08) and Codestral Embed ($0.15 per 1M)
  • Mistral Embed ($0.10 per 1M)
  • Voxtral Transcribe 3 (from $0.0035 per minute) and Voxtral TTS 3 ($13 per 1M characters)
  • OCR 4.1, OCR 3, Moderation 2

Recently Retired

  • Mistral Large 2.0 (128K): retired 2025-03-30
  • Mistral 7B, Mixtral 8x7B and 8x22B: retired 2025-03-30
  • Mistral Large 2.1: retired 2026-05-31
  • Devstral 2 and Magistral 1.2: retired 2026-07-31
  • Mistral Medium 3 and 3.1: retired 2026-08-31
  • Mistral Medium 3.5: deprecated, retires 2027-04-30 (move to Large 4)

Products

  • Mistral Studio: developer console, usage and cost tracking
  • Mistral Vibe: coding agent on web, CLI, VS Code and JetBrains, mobile
  • Le Chat: assistant with Free, Pro, Team and Enterprise plans
  • Enterprise: SAML SSO, audit logs, custom deployments

Real-World Usage

Funding and Scale

  • Series D: EUR 3B at a post-money valuation above EUR 21B, led by Samsung (September 2026)
  • Studio relaunched for enterprise agents and workflows in November 2026
  • Vibe 2.0 shipped in January 2026

User Feedback

  • G2: about 4.3 out of 5 from a small review base
  • Praised: pricing, speed, self-host option
  • Criticized: complex-task quality, support response

How It Compares

Feature Mistral AI Google DeepMind Alibaba Qwen
Flagship API model Mistral Large 4 Gemini 4 qwen-max
Price per 1M (in / out) $0.70 / $2.10 $2 / $10 $1.60 / $6.40
Flagship context 1M 1M Varies by model (open Qwen3.8-27B: 262K native)
Flagship weights Published (Ecosystem license) API only (Gemma is separate) API; open Qwen3.8 models separate

Sources: Mistral pricing · Large 4 card · Gemini pricing · Model Studio pricing

Is this your tool? Add a badge