New to this? Read it in simple words
- Mistral is a French AI company.
- Anyone can now download its biggest model, Large 4, and run it themselves.
- Mistral says it codes as well as the best open model from China.
- An independent tester puts it one point lower on coding, and further behind overall.
- Open weights
- The files of a trained AI model, shared so that anyone can download and run it.
- Parameter
- One of the many numbers inside an AI model that it learned in training.
- Licence
- The rules that say how something may be used.
On October 6, Mistral opened a preview of Large 4 and promised the weights by the end of the month. It kept that promise on October 29, after three more weeks of training, Mistral says.
What is released
The weights are on Hugging Face, so anyone can download the model and run it on their own hardware. Mistral also offers it through its API, and it says partners such as Nvidia, OpenRouter and Microsoft Foundry carry it.
The model card counts about 1 trillion parameters, with 49 billion active for each token, or 52 billion with some extra layers. The model can read up to 1 million tokens at once, and it has low, medium and high reasoning settings.
The licence is Mistral’s own, a modified MIT licence. Companies that sell access to the model as a service must first get a separate commercial licence from Mistral.
Artificial Analysis Intelligence Index. The leaders’ scores and the preview’s are as VentureBeat reports them.
Mistral’s claims and the outside test
Mistral says Large 4 scores 48% on the Artificial Analysis Coding Agent Index. It calls that “effectively tied” with GLM-5.3, the leading open model from China, while using 47% fewer tokens per task.
Artificial Analysis gives Large 4 48 points on that index and GLM-5.3 49, VentureBeat reports. On the broader Intelligence Index, Large 4 scores 26, up from 25 for the preview. The best open models, DeepSeek V4.1 Flash and GLM-5.3, score 32, according to VentureBeat.
Running the index cost $2.06 per task with Large 4, Artificial Analysis says. VentureBeat reports that this is 63% less than the preview cost.
Why Mistral says open matters
Mistral pitches the model to companies and governments that want control over their AI. Because the weights are open, access “can’t be cut off”, Pierre Stock, Mistral’s vice president of science, told TechCrunch.
“Open source is essential to ensure state-of-the-art AI does not end up concentrated in the hands of a few actors,” chief executive Arthur Mensch said, as SiliconANGLE reports.
A version tuned for cybersecurity work goes only to vetted partners, such as Orange Cyberdefense, under contracts. Mistral says it is adjusting the model’s safeguards, because tests showed it might help people carry out real attacks.
Sources
Every fact in this story comes from the sources below. Open them to check our work.
- 1
- 2Primary source · October 29, 2026Model card: mistralai/Mistral-Large-4-1T-A52B Mistral AI, on Hugging Face
- 3Research · Accessed October 31, 2026Mistral Large 4 (High) - Intelligence, Performance & Price Analysis Artificial Analysis
- 4Research · October 29, 2026Mistral Large 4 ‘Le Chonk’ arrives as the top open model outside China with improved coding performance and 63% lower cost-per-task VentureBeat
- 5Research · October 29, 2026Mistral releases weights for its trillion-parameter AI model, targeting cybersecurity TechCrunch
- 6Research · October 29, 2026Mistral releases open weights for Mistral Large 4, providing sovereign control for the company’s ‘Le Chonk’ AI model SiliconANGLE
The 48% coding score and the token saving are Mistral’s own claims. On two other coding tests, Mistral’s results are higher than those Artificial Analysis measured, so we leave them out.
Mistral releases the weights of its Large 4 model. Silicon AI News, October 31, 2026. https://siliconainews.com/stories/mistral-large-4-open-weights
Mistral released downloadable weights for its trillion-parameter Large 4, though it trails top Chinese open models and the licence has limits. How we rate the mood
The mood shows how the news frames the event for the tech and AI world: as an opportunity (6 and above) or as a loss (below 5). It is not our verdict on who is right.