IMPLICATOR
Pierre Stock, Mistral's vice president of science, called it "a pivotal moment in the global AI leadership race" when he told reporters about the release at a press briefing. The French lab on Thursday released the weights for Mistral Large 4, a trillion-parameter model it calls Le Chonk, a French wordplay on the company's cat mascot.
Mistral says the model ranks as the best open-weight model for coding. Its launch post puts the model effectively tied with China's GLM-5.3 on the Artificial Analysis Coding Agent Index.
Artificial Analysis ranks it second.
The evaluator's assessment supports Mistral on efficiency and lower costs. It also puts boundaries around the company's leadership claims.
What Changed
- Mistral released the weights for Mistral Large 4, a trillion-parameter model it calls Le Chonk.
- Artificial Analysis ranks it second among open-weight entries on its Coding Agent Index, one point behind GLM-5.3, with about 47% fewer tokens per task.
- Its cost per Intelligence Index task fell to $2.06 from $5.54 for the preview, a 63% reduction, though DeepSeek and MiMo score higher at lower cost.
- Mistral's own post says it sees a potential for uplift to conduct real attacks and is modifying its guardrails.
AI-generated summary, reviewed by an editor. More on our AI guidelines.
One point behind
On the Coding Agent Index, the model sits one point behind GLM-5.3 and two above DeepSeek. The reported one-point difference alone does not establish statistical significance.
Mistral's model used 8.5 million tokens per task, about 47% less than GLM.
One of Mistral's other coding numbers comes out lower in the outside test. Mistral reports 41% on Terminal-Bench 4.0, up from a self-reported 28% for the preview. Artificial Analysis's Intelligence Index assessment reports 25% on that test. The figures come from different reported evaluation contexts.
FREE WEEKDAY MORNING BRIEFING
Track the open-model race as scores come in.
The Implicator Morning Briefing filters the AI news cycle to the stories worth your attention and explains their consequences. From San Francisco, every weekday at 4:45 a.m. Pacific, 7:45 a.m. Eastern.
About five minutes. No hype. No spam.
On the broader Intelligence Index v5, the model scores 26, up from 25 for the preview. That makes it the highest-scoring open-weight model developed outside China, one point above Muse Spark Open. DeepSeek V4.1 Flash 1017 and GLM-5.3 lead the open-weight group at 32.
A smaller bill
The larger change is economic. Artificial Analysis reports that the updated model costs $2.06 per Intelligence Index task, down from $5.54 for the preview, a 63% reduction. Mistral's API rates are now $0.70 per million input tokens and $2.10 per million output tokens, roughly half the earlier $1.36 and $4.18.
DeepSeek and MiMo achieve higher overall Intelligence Index scores at substantially lower measured task costs. These benchmark costs do not measure the expense of self-hosting, customization, operational oversight or human review.
Mistral trained the model on 3,800 Nvidia Grace Blackwell GPUs in its European data centers. Arthur Mensch, its co-founder and chief executive, framed the funding gap in one line: "Other labs have raised hundreds of billions to develop equivalent capabilities; in three years, Mistral raised roughly $6 billion." Anthropic was valued at $965 billion as of August.
What customers ask for
Marjorie Janiewicz, Mistral's chief revenue officer, described customers seeking models they can retain and customize in an in-person interview with VentureBeat earlier this week.
"I don't want to get disconnected. I want to own my model," Janiewicz said, describing that customer demand.
Asked for business results attributable to the new model, she said systems incorporating it were still being built and that meaningful operational measurements would take longer.
Know someone who'd find this useful? ✉️ Email it to a friend in one click, or they can subscribe free here.
The weights ship under a modified MIT license. Businesses that operate a model-as-a-service offering must obtain a separate commercial license from Mistral before using the model commercially.
The cyber argument
Mistral also sells the model for cyber defense. Artificial Analysis confirms a Cyber Index score of 51, third among current open-weight models, behind DeepSeek V4.1 Flash 1017 at 57 and MiMo-V2.6-Pro at 56.
Stock argues that open weights protect buyers from losing access. "AI is going to be like electricity," he said. "Once you open source the weights, they're all over the internet. It's impossible to cut off access."
He also questioned whether Chinese open models can be trusted. "You have no guarantee that [Chinese models] won't behave maliciously. They can be threat vectors," Stock said.
Mistral's own launch post carries a warning. After monitoring cybersecurity-related use in live traffic since the preview, the company wrote that "we see a potential for uplift to conduct real attacks." It says it is modifying its guardrails.
Rafe Pilling, senior director of threat intelligence at Sophos, sees the trade-off in anything freely downloadable. "It's clearly a dual-use technology," he told Wired. Pilling said all advanced models are capable of going rogue without careful supervision, no matter their country of origin.
"It's about having the right monitoring and controls in place to be confident that, even if the model was trying to be deceptive or nefarious, you might have some way of detecting it," he said.
Frequently Asked Questions
What did Mistral release?
Mistral released the weights for Mistral Large 4, a trillion-parameter model nicknamed Le Chonk. The weights ship under a modified MIT license, and businesses that operate a model-as-a-service offering must obtain a separate commercial license from Mistral before using the model commercially.
Is Mistral Large 4 the best open-weight coding model?
Mistral says so. Artificial Analysis ranks it second on its Coding Agent Index, one point behind GLM-5.3. The reported one-point difference alone does not establish statistical significance.
How does it compare overall with Chinese open models?
On Intelligence Index v5 it scores 26, the highest among open-weight models developed outside China. DeepSeek V4.1 Flash 1017 and GLM-5.3 lead the open-weight group at 32.
How much cheaper is the final model?
Artificial Analysis reports $2.06 per Intelligence Index task, down from $5.54 for the preview, a 63% reduction. API rates are $0.70 per million input tokens and $2.10 per million output tokens. These costs do not include self-hosting or human review.
What do security experts say about releasing the weights?
Rafe Pilling of Sophos called it clearly a dual-use technology. He said all advanced models are capable of going rogue without careful supervision, no matter their country of origin.
AI-generated summary, reviewed by an editor. More on our AI guidelines.
Free AI briefing · Weekdays
The AI stories that matter, sourced and explained.
Join the Morning Briefing. It goes out every weekday at 4:45 a.m. Pacific, with later sends for the East Coast, Berlin and Tokyo.
Free when you sign up: The Paperclip Compendium, our tested guide to running AI agents.
Free. Unsubscribe in one click.
Marcus Schuler
Editor-in-Chief and founder of Implicator.ai. Former ARD correspondent and senior broadcast journalist with 10+ years covering tech. Writes daily briefings on policy and market developments. Based in San Francisco. E-mail: editor@implicator.ai
