Mistral releases Large 4 open-weight model with one trillion parameters
8 sources · The Decoder · qz.com · UA.NEWS- Boom: Mistral Large 4 has one trillion parameters, the company's largest model to date
- Doom: Model scores well on Intelligence Index but still trails Claude, GPT-6 and Chinese rivals
- Boom: Large 4 is planned for open-weights release, giving users direct control
- Neutral: Model was trained on Mistral's own European infrastructure
- Boom: Mistral pitches cybersecurity work that closed US models reportedly refuse to perform
The story in full
French AI company Mistral released Large 4 on October 6, 2026, a text-output model with one trillion parameters trained on the company's own European infrastructure. The model is planned for open-weights release and carries cybersecurity capabilities that Mistral says closed US models decline to support.
In the independent Intelligence Index, Mistral Large 4 scores notably higher than previous Mistral models but still trails Claude, GPT-6, and leading Chinese competitors. Mistral is positioning the model as a tool businesses can control directly, with cybersecurity testing as a primary use case distinguishing it from US rivals.
Analysis
429 wordsOn October 6, 2026, French AI company Mistral released Large 4, internally nicknamed Le Chonk, a text-output model with one trillion parameters, making it the largest model the company has built to date. Mistral trained the model on its own European infrastructure and is planning an open-weights release, meaning users and businesses would be able to download and run the model directly rather than accessing it through an API under a provider's terms. The company is pitching Large 4 specifically for cybersecurity use cases, arguing that closed US models decline to assist with that kind of work, which Mistral frames as a practical differentiator.
The release matters for a few reasons beyond the parameter count. Mistral is one of the few non-US, non-Chinese companies attempting to compete at the frontier, and the open-weights framing carries weight in the European policy environment, where sovereign AI infrastructure has become a priority. At the same time, the independent Intelligence Index places Large 4 behind Claude, GPT-6, and leading Chinese models, which complicates Mistral's positioning. The cybersecurity angle is also genuinely in dispute: whether frontier-level offensive security capability in an open-weight model is a feature or a risk is a question the field has not settled, and regulators have not fully addressed.
Pro-AI voices greeted the release warmly. Accounts like aennassiri called for more compute and framed the release as just a first step, while danslo pointed to the open-weight format, European origin, and cybersecurity performance as straightforwardly positive. rglover described it as the AI race moment they had been waiting for, and Euronews en Français noted Mistral's explicit claim to European sovereignty in the US-China competition. The lone anti-AI reaction, from Jakob S, was pointed: the argument is that Mistral is being outperformed on both intelligence and cost by Deepseek v4.1 Flash, making the release a losing position rather than a milestone. The middle ground is more measured. Ellen teapot and A.V. both described Large 4 as roughly competitive with Chinese open-weight models but meaningfully behind US frontier models, and theturtletalks called the gap to Chinese open models disappointing. Briefing Block suggested that multilingual document tests next week could reveal enterprise strengths that aggregate coding benchmarks miss, pointing to uncertainty about where the model actually shines.
The clearest thing to watch is the actual open-weights release itself, since several middle-ground voices noted the weights had not yet dropped at announcement. Independent benchmark runs on enterprise tasks, and any regulatory response to the cybersecurity capabilities Mistral is advertising, would do more to settle the competitive and policy questions than the launch claims alone.
What Pro-AI voices are sayingMistral Large 4 is celebrated as a major leap for European open-weight AI, with enthusiasm for its scale, cybersecurity features, and potential to set the pace in open-source development. Some see it as proof that Europe can compete with China.
Quote 1 of 13What Anti-AI voices are sayingSkeptics question whether the model genuinely beats Chinese rivals or merely cherry-picks benchmarks, and one voice raises concerns about removing human oversight in the review process.
Quote 1 of 5What Middle Ground voices are sayingMost see Mistral Large 4 as competitive with Chinese open-weight models but still behind US frontier models, with several noting benchmark claims need real-world validation before drawing firm conclusions.
Quote 1 of 13Add your take
0 reader votesSign in with Google to pick a side and post. Your vote moves the story's Doom / Boom score.
More Pro-AI reactions (12)
“Excited to see this! Nice that they are saying this is just a first step. Give them more compute!”
aennassiri, Hacker News · 14:24 UTC“Open weight, European, competes with GLM-5.3 on cybersecurity. What's not to like?”
danslo, Hacker News · 14:15 UTC“Very nice progress. Also I respect putting Kimi on those charts. Regardless of if they are beating the Pareto frontier (not now), model diversity is a good thing for humanity — I’m hopeful for the team to keep increasing their”
vessenes, Hacker News · 14:01 UTC“Welcome back, Mistral! Mistral 4 is here, a 1t, 49b active MOE with native vision and fully open weight.”
Joshua White, Bluesky · 13:18 UTC“instant new high score on felony-bench for that name, truly incredible”
Laurens, Bluesky · 14:42 UTC“MISTRAL having fun announcing their latest open weights model, Mistral Large 4 aka Le Chonk”
alina, Bluesky · 13:14 UTC“we fucking memed a frontier model into existence”
mlf. ⎔, Bluesky · 13:45 UTC“actually very excited to use this, the previous mistral models were too far behind in capabilities imo to for serious day-to-day use, but this is good enough that I'm actually very interested in how the model feels”
Laurens, Bluesky · 14:43 UTC“« On a accompli des miracles » : Mistral annonce Large 4, sa nouvelle IA surpuissante qui défie les cadors chinois”
Frandroid, Bluesky · 13:05 UTC“Mistral Large 4 est officiel : la France rattrape enfin la Chine en proposant l’un des meilleurs modèles d’IA ouverts du monde entier ➡️ www.presse-citron.net/mistral-larg...”
Presse-citron, Bluesky · 14:43 UTC“claiming enhanced performance and security capabilities for clients seeking to keep their data private instead of relying on closed models from the likes of Anthropic and OpenAI”
AFP News Agency, Bluesky · 15:00 UTC“Mistral 🤖 hat zumindest das Internet verstanden”
Michael Horak aka fatmike182, Bluesky · 14:12 UTC
More Anti-AI reactions (4)
“congratulations to mistral for releasing the first pareto roadside-ditch model”
samhain ⎔, Bluesky, skeptic · 14:51 UTC“The french are getting mogged, both on intelligence and cost, by Deepseek v4.1 Flash”
Jakob S, Bluesky, skeptic · 13:42 UTC“"an internal research model achieved this cool thing; we would like to offer you access to a model that is not that one" feels like the vacuum cleaner salesmen who demonstrate how good the machine is, but then have to”
Matthew Prorok, Bluesky, skeptic · 11:49 UTC“The review seat was the one place a human signature was structural - hand it to an unnamed system and provenance doesn't weaken, it exits.”
seldondance.bsky.social, Bluesky · 09:42 UTC
More Middle Ground reactions (12)
“Weights currently closed but promise to open source them by the end of the month. Great release movie.”
spwa4, Hacker News · 13:36 UTC“With the release of its new trillion-parameter model, Mistral is hoping to demonstrate it’s “still in the race” to build frontier-level artificial intelligence.”
WIRED, Bluesky · 13:20 UTC“... without specifying which Chinese models or by what measure Mistral's model beat them ...”
codewright, Bluesky, skeptic · 11:24 UTC“it might be ... pretty decent? though this is still being debated amongst experts”
thalassa 🌻🔆, Bluesky · 13:30 UTC“Mistral will totally not just beat chinese models but all models including the ones by "Open"AI and Anthropic. /half kidding”
ChaoticHuman, Bluesky · 12:16 UTC“Mistral Large 4 looks at least vaguely competitive with Chinese open-weight models”
ellen teapot 🇨🇦🏳️⚧️, Bluesky · 14:15 UTC“the Le Chonk name is a good sign that Mistral is not purely and exclusively a compliance department for procurement requirements from brussels, but that the company has some of their own identity as well”
Laurens, Bluesky · 15:03 UTC“Next week, multilingual document tests could reveal enterprise strengths that aggregate coding rankings miss.”
Briefing Block, Bluesky · 13:49 UTC“The "matches Chinese frontier at cheaper compute" claim is worth watching closely: either it's real hardware efficiency, or it's a different benchmark.”
AI Founders ONLINE, Bluesky, skeptic · 10:46 UTC“Hard to see it displacing Qwen's coder as a daily driver until it survives real codebases.”
aichina.news, Bluesky, skeptic · 13:25 UTC“Competitive with Chinese open source models. Quite behind US frontier.”
A.V., Bluesky · 14:04 UTC“it does not sound very exciting to match Chinese models in some areas”
input, Bluesky, skeptic · 09:41 UTC
Sources
9 articles from 8 outlets- The DecoderMistral Large 4 is Europe's trillion-parameter answer to US models that refuse security work
- qz.comMistral launches Mistral Large 4 open-weight AI model
- UA.NEWSFrance’s Mistral announces open-weight Large 4 model — Channel NewsAsia
- Hacker News front page (AI)Mistral Large 4
- WSJMistral to Release New AI Model to Better Compete With U.S. Rivals
- FinimizeMistral’s New Open-Weight AI Model Leans Into Cyber Testing
- Hacker News front page (AI)Mistral Large 4
- zdnet.comMistral’s new Le Chonk model brings AI cybersecurity to your business – and you control it
- VentureBeatMistral debuts Large 4 ‘Le Chonk', a 1-trillion parameter text output model with high benchmarks planned for open weights release
Related Splits
All stories →Meta and Microsoft cut internal use of Anthropic's Claude
Meta and Microsoft have sharply reduced their internal use of Anthropic's Claude, according to a report by The Information published on October 5, 2026. Microsoft cut...
OpenAI sued over Astra trademark by company TradeSun
A company called TradeSun filed a trademark infringement lawsuit against OpenAI on or around October 5, 2026, over OpenAI's use of the name Astra for an...
Anthropic moving forward with IPO plans amid market uncertainty
Multiple outlets reported in late September and early October 2026 that Anthropic is preparing to go public, with one headline citing a figure of $518 billion...