Mistral AI

Mistral AI

Overview

Mistral AI emerged in 2023 as Europe's leading AI startup, founded by a team of researchers with backgrounds at Meta AI and DeepMind. The company has distinguished itself through a commitment to efficient, open-weight models and a balanced approach to open-source development with commercial viability. Operating from Paris with strong European backing, Mistral positions itself as a challenger to American AI incumbents while maintaining deep partnerships with leading semiconductor and infrastructure companies. The company's aggressive growth trajectory and €1B revenue target for 2026 reflect confidence in enterprise adoption of its models.

In Spring 2026, Mistral accelerated its model cadence with three notable releases: Mistral Small 4 (a hybrid model unifying instruct, reasoning, and coding in a single multimodal architecture with 256K context), Mistral Medium 3.5 (an open-weight frontier-class multimodal model with adjustable reasoning effort that scored 77.6% on SWE-Bench Verified), and continued iteration on Codestral and the Voxtral text-to-speech family. The Medium 3.5 release in particular pushed Mistral into the upper tier of agentic / coding model providers — open-weight, mid-priced, and ahead of Devstral 2 and Qwen3.5 397B A17B on SWE-Bench Verified.

Key Details

  • Founded: 2023
  • CEO: Arthur Mensch
  • Headquarters: Paris, France
  • Total Funding: €3.05B (7 funding rounds)
  • Latest Round: €2B (September 2025), led by ASML
  • Current Valuation: ~€13.7B (~$15B USD)
  • 2026 Revenue Target: €1B (~$1.2B USD)
  • Website: https://mistral.ai

Current Models

  • Mistral Medium 3.5 (mistral-medium-3-5) — Frontier-class multimodal mid-tier model with adjustable reasoning effort; 77.6% on SWE-Bench Verified (ahead of Devstral 2 and Qwen3.5 397B A17B); $1.50/$7.50 per million input/output tokens; Modified MIT open-weight license; released April 29, 2026
  • Mistral Small 4 (mistral-small-2603) — Hybrid model unifying instruct, reasoning, and coding in a single multimodal architecture; 256K context window (Spring 2026)
  • [[Mistral Large 3]] — Apache 2.0 licensed, 41B active parameters / 675B total MoE, 256K context window, $0.50/$1.50 per million tokens (input/output), released December 2, 2025
  • [[Mistral Medium 3]] — Mid-tier model (predecessor to Medium 3.5); 131K context window, $0.40/$2.00 per million tokens, released May 7, 2025
  • [[Codestral]] — Code-specialized model, 256K context window, support for 80+ programming languages, initially released May 2024 with 2025 updates
  • Voxtral TTS — Open-weight text-to-speech model released March 23, 2026; supports 9 languages (English, French, German, Spanish, Dutch, Portuguese, Italian, Hindi, Arabic); zero-shot voice cloning from as little as 3 seconds of reference audio; positioned as a direct challenger to ElevenLabs and OpenAI's voice stack
  • Forge — Mistral's enterprise training/customization platform introduced in the 2026 portfolio refresh
  • Magistral — Reasoning-focused model line (chain-of-thought; o-series competitor); now consolidated into Small 4 architecture
  • Pixtral — Multimodal vision family (document, chart, image understanding); now consolidated into Small 4 architecture
  • Devstral — Agentic-coding line; now consolidated into Small 4 architecture

Key People

No individuals from the "Tracked People" list are affiliated with Mistral AI.

Recent Developments

  • Mistral Medium 3.5 Launch (April 29, 2026): Open-weight (Modified MIT license) multimodal frontier-class mid-tier model with adjustable reasoning_effort parameter. Scored 77.6% on SWE-Bench Verified — ahead of Devstral 2 and Qwen3.5 397B A17B. Pricing: $1.50/$7.50 per million input/output tokens, positioning Medium 3.5 as Mistral's primary agentic / coding play.
  • Mistral Small 4 Launch (March 16, 2026): Hybrid instruct/reasoning/coding model with 256K context, multimodal input — explicitly unifies the previously separate Magistral (reasoning), Pixtral (multimodal vision), and Devstral (agentic coding) lines into a single architecture.
  • Voxtral TTS Launch (March 23, 2026): Open-source text-to-speech with 9-language coverage and 3-second zero-shot voice cloning, framed as a direct challenge to ElevenLabs and OpenAI's voice stack.
  • March 2026 Portfolio Sprint: Between March 16 and March 31, 2026, Mistral shipped the unified Small 4 model, Voxtral TTS, an open-weight formal-proof agent, an enterprise training platform (Forge), a developer CLI, and a founding role in NVIDIA's Nemotron Coalition — one of the densest two-week release windows of any AI lab in 2026.
  • €2B Funding (September 2025): Led by ASML, with NVIDIA and other strategic investors participating.
  • €1B Revenue Target for 2026: CEO Arthur Mensch confirmed at Davos.
  • €1.2B ($1.43B) Sweden Data Center Investment: New infrastructure opening in 2027.
  • Acquisitions Strategy: CEO Arthur Mensch actively exploring acquisitions to accelerate growth.

Why They Matter

Mistral AI represents the strongest established European AI player, providing credible alternatives to U.S. models while maintaining open-source principles. The April 2026 Medium 3.5 release is particularly important: at 77.6% on SWE-Bench Verified and $1.50/$7.50 pricing under a permissive license, Mistral has positioned itself as the open-weight default for enterprise coding and agentic workflows — a segment where DeepSeek V4 and Qwen 3.5 dominate from China but where Mistral has the data-residency, regulatory, and EU-investor advantages that European enterprises increasingly demand. The company's focus on efficiency and the substantial European investment in infrastructure signals a commitment to AI sovereignty and reduced dependence on American technology. Mistral now sits in a three-way EU competitive frame alongside the newly-emerged Ineffable Intelligence (DeepMind alumnus David Silver's London RL lab) and the broader U.K. Sovereign AI Fund's portfolio — making 2026 the year European frontier AI finally has multiple credible national-champion candidates.

Last Updated

May 8, 2026

→ Back to Companies