Mistral unveils Le Chonk: a 1T-parameter open-weights model
The biggest open-weight release outside China lands in public preview today, and the country that spent the week promising its own frontier model just put a price on the ambition.
Mistral has opened a public preview of Mistral Large 4 — codenamed "le Chonk" — a 1 trillion-parameter mixture-of-experts model with 49 billion active parameters, natively multimodal, which the company calls its largest and most capable model to date. The preview API is live today on Mistral Studio at $1.36 per million input tokens and $4.18 per million output tokens, and Mistral says the weights drop by the end of the month; TNW reports the release date as October 27. The benchmark claims are Mistral's own, with several coding scores coming from Artificial Analysis evaluations run privately ahead of the harness's public launch: 61.7% on DeepSWE v1.1, 28.3% on Terminal-Bench 4.0, and a combined Coding Agent Index score of 49.8% that puts it ahead of DeepSeek V4 Pro 0813 and Qwen3.8 Max. Its strongest independent-looking number is cybersecurity — 82% on an Artificial Analysis test that asks a model to reproduce a real vulnerability in open-source software and then patch it, which Mistral says is the highest of any model, largely because closed models like Claude Opus 5.5 and GPT-6 Astra score near zero by refusing the task outright. Trained from scratch on 3,800 NVIDIA Grace Blackwell GPUs in Mistral's own European datacenters and funded by the €3 billion Series D it raised in September, ML4 is the clearest test yet of whether Europe's sovereignty bet can produce frontier-adjacent capability on open weights — while CNBC is right to note the model still lags the frontier in areas like coding, and worth remembering that until the weights are actually downloadable, "open-weight" is a promise, not a fact. Until release, Mistral is red-teaming the model with cybersecurity leaders, vetted partners and state authorities, who get the same model with reduced moderation and expanded cyber capabilities.
South Korea has proposed $3.49 billion (4.7 trillion won) in its 2027 budget to build a homegrown frontier AI model — a nine-fold step up from its previous program, pending National Assembly approval expected in December. The equity investment sits inside a 9.4 trillion won AI budget request, with the new frontier competition opening after the vote and the program potentially starting as early as March. The ambition is calibrated: Vice Minister Ryu Je-myung said going head-to-head with the biggest US labs is "unrealistic," but a model competitive with leading Chinese open models is within reach, backed by roughly 29,000 government GPUs to be allocated by year-end.
What to watch: Mistral's end-of-month weights drop and whether independent evaluations confirm the self-reported scores, and Korea's December assembly vote on the frontier-model fund.
Is open weights from a €21 billion European lab the real counterweight to China's model exports — or does it only count once the files are downloadable? Tell us in the comments.
