AI Change Tracker
4

Mistral ships open weights for Mistral Large 4 ("Le Chonk"), a 1.05T-parameter model aimed at agentic coding and cybersecurity

2026-10-29

Mistral published open weights for Mistral Large 4 ("Le Chonk"), a 1.05T total / 52B active parameter MoE model with a 1M-token context window and text+image input, under its MEOW license, which requires a separate commercial license for model-as-a-service businesses. Pricing is $0.70/$2.10 per million input/output tokens, about 50% below the earlier preview. Mistral says the model ranks #1 among open models on Cognition's FrontierCode benchmark and was tested for cyber-defense use by CrowdStrike, Cloudflare, and Zscaler; a Mistral researcher reported a 64% human-preference win rate against GLM 5.3. Artificial Analysis independently confirmed the parameter counts and context window.

Significance 4: A trillion-parameter open-weights release from the leading non-Chinese open lab, with independent confirmation of specs, is a significant new entrant per the rubric; capped below 5 since the win-rate and FrontierCode-#1 claims are lab/vendor-reported and unverified.

technology financial Open weights Release Pricing change

Implications · machine-drafted, not owner judgment

A second actually-shipped (not just announced) non-Chinese open-weights frontier model this month extends the open field beyond the Chinese labs that have dominated recent open releases — relevant to the open-weights-one-generation-behind thesis's geographic diversity. The MEOW license's carve-out against commercial model-as-a-service resale is a notable middle path worth watching as other labs navigate open-weights monetization.

Watch for
  • Independent (non-Mistral) confirmation of the FrontierCode #1 and GLM-5.3 win-rate claims
  • Whether a cloud provider tests the MEOW license's commercial-use restriction

Sources

← Back to feed