Recommended Free Tools
Mistral announced Mistral Large 4 on October 6, 2026, but it has not yet released the model weights. The model is available in public preview through Mistral’s API; Mistral says weights are planned by the end of October, and its Hugging Face listing currently estimates October 31. That makes “open-weight” a description of the planned release, not a way to download or run the model today.
What is Mistral Large 4?
Mistral describes Large 4, nicknamed “Le Chonk,” as a general-purpose, natively multimodal Mixture-of-Experts (MoE) model combining instruction following, reasoning and agentic capabilities. MoE models use different parts of a larger network for different inputs or tasks; the total parameter count therefore is not the same as the number of parameters active for every request.
As an Amazon Associate I earn from qualifying purchases.
Mistral’s announcement rounds the model to 1 trillion parameters. Its documentation gives a more precise total of 1.05 trillion parameters and lists a 1.6-billion-parameter vision encoder. The active-parameter figures also differ by counting method: the hosted model listing says 49 billion when embeddings and output layers are excluded, while the documentation gives 52 billion including them. These figures describe different scopes, rather than a contradiction about one identical count.
Free tools Windows power users keep installed
One-click scans. No signup required.
Is Mistral Large 4 open-weight, and can you download it?
Not yet, as of October 7, 2026. Mistral’s October 6 announcement opened a public preview API and said the weights would be released by the end of October. Mistral’s hosted Hugging Face listing gives October 31, 2026 as the current estimated release date. That is an estimate, not confirmation that the files will be available on that date.
#1 Best Overall
| Access route | Status | What it means |
|---|---|---|
| Public preview API | Available at announcement | Use the model through Mistral’s hosted service; you do not download or operate the weights yourself. |
| Downloadable weights | Planned for later in October; listing estimates October 31 | Could enable self-managed use after release, but the final files, license, and operating requirements were not established in the available announcement and listing. |
“Open-weight” does not necessarily mean open source under every definition or permit unrestricted commercial use. Until Mistral publishes the final license, the permitted uses and restrictions cannot be stated. The current preview is not evidence that local deployment is available.
What does the model support?
Mistral’s documentation specifies a 1-million-token context window and API support for structured outputs, function calling, document question-answering, batching, and agent or conversation use. A large context window can let a request include more material, but it does not by itself guarantee accurate answers or reliable task completion.
Mistral says Large 4 was trained from scratch on 3,800 NVIDIA Grace Blackwell GPUs in its European data centers and that the public preview is served on that infrastructure. The company also says a significant share of its training data was multilingual, covering more than 160 languages, including every official language of the European Union. These are statements from Mistral, not independently verified infrastructure or training-data findings.
How strong is Mistral Large 4?
Mistral reports results in coding, agentic workflows, multimodal understanding, cybersecurity, finance, and law. Its announcement cites 82% on a vulnerability reproduction-and-patching test and 93% on Cybench, as well as internal human evaluation for coding and comparisons involving CAD, STEM, finance, and coding. Those are company-reported figures; the announcement’s claims should not be read as independent benchmark confirmation.
Le Monde reported on October 6 that the claims had not yet been confirmed in regularly updated independent rankings. It also described preliminary company figures for Deep SWE 1.1 and summarized Mistral’s comparative claims in finance, cybersecurity, and geospatial analysis. Until independent results are available, the announced scores are useful as a description of what Mistral says the model can do, not proof that it outperforms a particular competitor in typical use.
What does Mistral Large 4 cost to use?
At the time of the October 6 announcement, Mistral’s announcement page displayed API rates of $1.36 per million input tokens and $4.18 per million output tokens. Mistral’s documentation page displayed different rates: $0.68 per million input tokens, $0.07 per million cached input tokens, and $2.09 per million output tokens. Because the two official pages showed conflicting prices, neither set should be treated as a settled current rate; check Mistral’s live pricing information before estimating costs. These are API token prices, not a price for downloading weights.
Rank #4
What is still unknown about the weights?
The planned weight release may change how people access and deploy Large 4, but the available materials do not settle several practical details:
- The final license and its commercial-use conditions.
- The exact model files and any release-day deployment guidance.
- The hardware and memory needed for local inference, which depend on the released artifact and deployment setup.
- Whether the estimated October 31 date will hold.
Until those details are published, it is premature to promise that an individual computer can run the model or to give a definitive self-hosting cost or hardware recommendation.
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.
