- Mistral has released open weights for Large 4, nicknamed “Le Chonk,” a roughly 1-trillion-parameter multimodal model, after first unveiling it earlier in October 2026.
- Mistral says Large 4 leads open-weight benchmarks in coding and defensive cybersecurity tasks and offers three reasoning modes to balance cost against performance.
- The company claims the model uses 47% fewer tokens per task than Zhipu’s GLM-5.3, though that efficiency figure has not been independently verified.
- The weights are available via Hugging Face and Mistral’s API, letting businesses and governments self-host the model instead of relying solely on Mistral’s hosted service.
What Happened
Mistral released open weights for Large 4, a roughly 1-trillion-parameter multimodal model nicknamed “Le Chonk,” making it available via Hugging Face and the company’s own API for businesses and governments to deploy directly, following the model’s initial unveiling earlier in October 2026, Mistral said in its release announcement.
Why It Matters
The release continues Mistral’s pattern of open-sourcing flagship models even as it raises capital at rising valuations, including a reported push toward a EUR20 billion valuation. It also sharpens the three-way race among open-weight labs that already includes DeepSeek and Meta’s Llama line (see our Mistral vs. Llama 4 vs. DeepSeek V4 comparison).
Technical Details
Mistral claims Large 4 leads open-weight benchmarks in coding and defensive cybersecurity tasks and offers three distinct reasoning modes that let users trade cost against performance. The company says the model needs 47% fewer tokens per task than Zhipu’s GLM-5.3 — a cost-efficiency claim Mistral has made publicly but that has not yet been independently verified by third-party benchmarks.
Who’s Affected
Enterprises and government agencies seeking self-hostable alternatives to closed frontier models; competing open-weight labs such as DeepSeek and Meta, whose models Mistral directly benchmarks itself against; and cybersecurity teams Mistral is specifically targeting with the release’s defensive-security framing.
What’s Next
Independent benchmark results against GLM-5.3 and other open-weight rivals are likely to surface now that the weights are public. Mistral has not disclosed a timeline for enterprise support contracts tied to self-hosted deployments of Large 4.
