France’s Mistral AI announced Mistral Large 4 on October 6, 2026. Also called “Le Chonk,” it is described in Mistral’s catalog as an open-weight, general-purpose multimodal model. Initial access was reported through a moderated API; Mistral planned to release the model weights on October 27, after further training and safety testing. The announcement’s performance claims remain preliminary, not independently confirmed rankings.
What is Mistral Large 4?
Mistral Large 4 is Mistral AI’s new general-purpose multimodal model, listed in the company’s catalog as version 26.10. “Frances Mistral” is not the company name: the announcement concerns France-based Mistral AI. The nickname “Le Chonk” refers to the model’s reported scale.
As an Amazon Associate I earn from qualifying purchases.
Axios reported that the model has 1 trillion parameters, with 49 billion active parameters, and was trained on 4,000 Nvidia Grace Blackwell GPUs over two months in Mistral’s European data centers. These are reported specifications and training details, not independently verified measurements. Mistral’s catalog characterizes its generalist models as suited to broad reasoning, coding, tool use, and agentic tasks; that description does not establish a particular local hardware requirement.
When will Mistral Large 4 be available?
At launch, Axios reported that access was initially through a moderated API. Mistral planned an open-weight release for October 27, 2026, following additional reinforcement learning and safety testing; Le Monde also reported that security testing was ongoing. Those dates and access arrangements were plans reported on October 6, not confirmation that the weights are now available. Check Mistral’s current catalog and access terms before relying on them.
#1 Best Overall
Is Mistral Large 4 open source?
Mistral calls Large 4 open-weight. That means the model weights are intended to be made available, which can give developers more flexibility to adapt or deploy a model than API-only access. “Open-weight” should not automatically be read as a claim that every component, training detail, or term of use is unrestricted; the applicable license and release materials determine what users may do.
Weights can also make downstream use harder for the developer to control: others may modify or deploy a model outside the original provider’s safeguards. Mistral’s VP of science Pierre Stock argued that releasing models could accelerate cyber defense, saying that open models, including Large 4, could help “accelerate the defense part way more.” That is an argument about potential benefit, not independent evidence that release makes the model safe.
What can the model do, according to Mistral?
Mistral presented Large 4 as relevant to long coding tasks, finance and spreadsheet work, cybersecurity, geospatial image analysis, and industrial design and production. These are company-reported areas of strength or intended use, not independently validated performance guarantees. The multimodal label indicates that the model is designed to work across more than one kind of input or output; it does not, by itself, specify every supported format or capability.
How strong is Mistral Large 4?
The performance picture is still provisional. Le Monde reported Mistral’s preliminary score of 63% on Deep SWE 1.1 for long coding tasks. In that report, the result was described as roughly on par with GLM 5.3, while leading models reached 74%. These are dated, reported comparisons, not a settled head-to-head evaluation across identical conditions.
Mistral also said Large 4 matched leading open models on some specialized finance, cybersecurity, and geospatial tasks. Le Monde attributed “narrows the gap” to Mistral co-founder Guillaume Lample, but noted that the company’s claims had not yet been confirmed in regularly updated independent rankings. Pierre Stock likewise acknowledged, “We’re not there yet on the frontier,” in reference to leading closed models.
How should you compare it with other AI models?
A single “best model” label would overstate what is established so far. A useful comparison depends on the job and on evidence beyond the launch announcement.
Rank #4
- Access: Large 4 was initially reported as API-only with moderation, with weights planned later. Availability and terms can change.
- Deployment and control: An open-weight release may allow adaptation and self-deployment, subject to its license and practical infrastructure needs. API access leaves hosting with the provider.
- Task performance: Treat the reported coding score and specialized-task claims as preliminary until comparable independent evaluations are available.
- Modality and fit: The catalog describes Large 4 as multimodal and general-purpose, but users should verify support for their specific inputs and workflow.
- Safety and cost: Release controls and inference or training costs matter in deployment decisions. The launch reporting does not provide a complete apples-to-apples cost comparison.
Does the announcement mean you need new hardware?
No specific consumer device or hardware purchase follows from the announcement. The model is software, and the cited descriptions do not set a local hardware specification. Users accessing the API rely on the provider’s infrastructure; self-hosting decisions depend on the eventual model files, license, and deployment requirements.
Quick wins for a faster PC:
Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Repair Windows errors before they cause bigger problemsFix Now →Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
