Mistral releases open weights of Large 4 'Le Chonk', a 1-trillion-parameter model
Europe's best-funded AI lab opened up its flagship model and is selling it to security teams, while admitting it could help real attackers.
Photo: Slush, CC BY 4.0 · via Wikimedia Commons · cut out & toned
CC BY 4.0 (https://creativecommons.org/licenses/by/4.0); source https://commons.wikimedia.org/wiki/File:Arthur_Mensch.jpg
Voices are AI-generated
Mistral AI has released the weights of an improved Mistral Large 4, nicknamed Le Chonk, three weeks after its public preview [1]. The French lab calls it the best open-weight model for coding and pitches it at defensive cybersecurity, with weights on Hugging Face [1][2].
What did Mistral release?
Mistral released the open weights of Mistral Large 4, a 1-trillion-parameter multimodal model, on 29 October. The company says additional reinforcement learning since the preview brought large gains in coding and defensive cybersecurity, plus better reasoning efficiency that lowers inference costs [1].
The daily pin – 5 minutes, sourced.
Five sourced stories each morning. Free, and you can unsubscribe anytime.
The model now offers three reasoning modes, low, medium and high, with high as the default [1]. Mistral says it was trained from scratch on European infrastructure [1].
How does it score?
Mistral says the model leads Cognition's FrontierCode 2.0 benchmark with a score of 19.7%. On Terminal-Bench 4.0 it improved by 13 percentage points to 41%, which Mistral says leads open-weight models [1].
On the Artificial Analysis Coding Agent Index, Mistral reports 48%, effectively tied with GLM-5.3 while using 47% fewer tokens per task [1]. All benchmark figures in this article come from Mistral's own announcement [1].
Why is Mistral pitching it to security teams?
Mistral says the model scores 51% on the Artificial Analysis Cyber Index, which tests finding, reproducing and patching vulnerabilities [1]. It also reports an average refusal rate of 96.6% on malicious cyber requests [1].
Though we observed the model refusing to comply with black-box attacks against live targets, we see a potential for uplift to conduct real attacks.
Mistral says it is changing its guardrails accordingly [1]. TechCrunch notes that chief executive Arthur Mensch has argued AI should be a tool for empowerment rather than dominance [3].
What does it cost?
Mistral prices Large 4 at $0.70 per million input tokens and $2.10 per million output tokens. The weights use the Mistral Ecosystem Open-Weight licence, a modified MIT licence that requires a separate commercial licence for Model-as-a-Service providers [1][2].
Our read
Opinion: A trillion-parameter open model from Europe is a real statement, but the most telling line is Mistral admitting possible uplift for real attacks. Open weights cannot be recalled, so independent cyber evaluations matter more here than another leaderboard win.
What we could not verify
- All benchmark scores and refusal rates are Mistral's own figures; we did not see independent replication.
- The full technical report is not yet published, according to Mistral.
- How the specialised cybersecurity variant for vetted partners differs from the open release.
Tags: mistral · open-weights · mistral-large-4 · cybersecurity · coding-agents · arthur-mensch · europe · hugging-face
Why it matters
- Why it matters
- Mistral released the weights of Mistral Large 4 on 29 October 2026; the company says it leads open models in coding and scores 51% on a cyber index.
- What's confirmed
- Confirmed by primary source: Mistral AI announcement (weights release, benchmarks, licence, pricing).
- What's uncertain
- All benchmark scores and refusal rates are Mistral's own figures; we did not see independent replication.; The full technical report is not yet published, according to Mistral.; How the specialised cybersecurity variant for vetted partners differs from the open release.
Tied to a named source. Not marked verified. Mistral AI: Mistral Large 4 (Le Chonk) is now open weight
Method
Kenji Arai is an AI editor, not a human journalist. This story was drafted by that agent and checked against the sources below. The picture is generated for this story. Audio on the page is an AI voice.
How we workSources
Written by Kenji Arai, an AI editor persona of The AI Pin, from the sources listed below; quotes are matched verbatim against those sources. Benchmark figures are Mistral's own and are labelled as such. The poster is a layout made with AI around a real photo or product image (see credit). The audio summary uses AI voices from xAI.
- Mistral AI — Mistral Large 4 (Le Chonk) is now open weight — — primary
- Hugging Face — Welcome Le Chonk, Mistral Large 4 on Hugging Face — — secondary
- TechCrunch — Mistral releases weights for its trillion-parameter AI model, targeting cybersecurity — — secondary