Mistral Le Chonk, also called Mistral Large 4, is a reported 1-trillion-parameter open-weight model that uses a mixture-of-experts design, activating about 49 billion parameters per task. As of October 17, 2026, it was available through a preview API, while downloadable weights were still pending and early reported benchmarks had not shown it surpassing top closed models.
Key takeaways
- Le Chonk is also known as Mistral Large 4 and is positioned by Mistral as an open-weight alternative to leading closed AI models.
- The model is reported to have 1 trillion parameters in total, with about 49 billion activated for a given task through its mixture-of-experts design.
- Mistral highlights coding, cybersecurity, and other technical and business tasks as intended use cases.
- Early reported independent benchmarks suggest competitiveness with some open-weight models but do not establish that Le Chonk beats leading closed systems.
- As of October 17, 2026, users could test the model through a preview API; downloadable weights had not yet been released.
Mistral has introduced Le Chonk, also known as Mistral Large 4, as an open-weight model designed to compete with leading closed AI systems. The company says it can handle demanding technical and business tasks, including coding and cybersecurity. Early independent benchmarking, however, offers a more measured picture: Le Chonk appears competitive with some open-weight models, but has reportedly trailed top closed systems so far.
As of October 17, 2026, users could try the model through a preview API. Mistral had said downloadable weights would follow later in October, but they were not yet available according to the reporting reviewed for this article.
Mistral's open-weight model and its 1 trillion parameter design
Le Chonk is reported to contain 1 trillion parameters in total. It uses a mixture-of-experts design, which means the model routes each task through selected portions of its full network rather than activating every parameter for every response. Mistral reportedly activates about 49 billion parameters at a time.
That distinction matters: the headline parameter count describes the model's overall capacity, not how many parameters are used for each prompt. The mixture-of-experts approach is intended to bring together broad capacity and more selective computation. The reported scale also positions Mistral Large 4 as a substantial step in the company's effort to build models that can serve as alternatives to proprietary AI systems.
Mistral describes Le Chonk as open-weight. That means the model's weights are intended to be made available for others to download and use, though access depends on the release and its terms. It is not the same as saying every part of the model's development, training data, or infrastructure is open.
What Le Chonk is built to do - and how its benchmarks compare
Mistral's pitch is that Le Chonk can rival leading closed models across a range of complex work. The company highlights coding, cybersecurity, and other business and technical tasks. Those are areas where performance can depend on more than general knowledge: models need to follow instructions, reason through multi-step problems, and produce usable outputs.
That ambition should be separated from what early evaluations have established. Initial independent benchmarking reportedly placed Le Chonk behind leading closed models, while showing a more favorable comparison with some open-weight competitors. The available reporting does not establish a definitive overall ranking or show that it outperforms the top closed systems across the tasks Mistral names.
Benchmark results are also only one part of the picture. Performance can vary by task, evaluation method, and the version tested. For developers and businesses, practical considerations such as reliability, response speed, deployment options, and the final usage terms will matter alongside benchmark scores. More independent results will help clarify where the model is strongest.
Le Chonk availability: Preview API now, weights later
At the time of writing, Mistral Large 4 was accessible through a preview API, giving users a way to test it without downloading the model. The downloadable weights were planned for later in October, but had not yet been released, according to the supplied reporting. That means claims about running the model independently or adapting it locally should be treated as future possibilities until the weights are actually available and their terms are clear.
For now, Le Chonk is a notable model announcement, not a settled verdict on the competitive landscape. Its trillion-parameter scale and open-weight direction are significant, while the early evidence supports a more cautious conclusion than Mistral's headline ambition. Follow our coverage for updates when Mistral releases downloadable Le Chonk weights and more independent benchmark results.
By the numbers
1 trillion total parameters
The article's supplied reporting describes this as the reported total size of Mistral Le Chonk; it is a model specification, not an independently verified benchmark result.
About 49 billion parameters activated at a time
The supplied reporting attributes this figure to Le Chonk's mixture-of-experts design, which activates selected parts of the model for each task.
Preview API access was available as of October 17, 2026
This is the availability status stated in the article based on the reporting reviewed; downloadable weights were reported as planned for later in October but not yet released.
Step by step
- 01
Check the current release status
Confirm whether Mistral has released downloadable weights or updated the preview API since the October 17, 2026 status described in the article.
- 02
Review the model terms
Read the applicable access and usage terms before testing the API or planning a deployment, and verify what the eventual weight release permits.
- 03
Test representative tasks
Evaluate the model on coding, cybersecurity, or business tasks that reflect your actual workload, using consistent prompts and success criteria.
- 04
Compare results across models
Use the same test set to compare Le Chonk with relevant open-weight and closed models, and record quality, reliability, and response speed.
- 05
Validate deployment needs
Assess operational requirements and confirm that the model's release format and terms support your intended hosting, adaptation, or use.
Frequently asked questions
What is Mistral Le Chonk?
Mistral Le Chonk is the name used for Mistral Large 4, an open-weight model from Mistral. It is designed for demanding technical and business tasks, including coding and cybersecurity. Mistral presents it as a potential competitor to leading closed AI systems, but early reported benchmarks do not establish that it outperforms them.
How many parameters does Mistral Le Chonk have?
Mistral Le Chonk is reported to have 1 trillion total parameters and to activate about 49 billion at a time. Its mixture-of-experts design routes tasks through selected parts of the network rather than using every parameter for every response. These figures are reported specifications, not independent measurements in the article.
Is Mistral Le Chonk open source?
Mistral describes Le Chonk as open-weight, which is not necessarily the same as fully open source. The weights are intended to be downloadable, but access and permitted uses depend on the release terms. The article notes that the weights had not yet been released as of October 17, 2026.
How can I try Mistral Le Chonk?
As of October 17, 2026, users could try Mistral Le Chonk through a preview API. Check Mistral's current access information and terms before testing, since availability may change. Downloadable weights were planned for later in October but were not yet available according to the reporting reviewed.
Does Mistral Le Chonk outperform top closed AI models?
Early reported independent benchmarking had not established that Le Chonk outperforms top closed AI models. The initial picture suggested it was competitive with some open-weight models but trailed leading closed systems so far. Results can vary by task, evaluation method, and model version, so additional independent testing is needed.
