Google's open multimodal embedding model that maps text, images, audio and video into one vector space, under Apache 2.0
Mistral Large 4
New· 4Mistral AI's one-trillion-parameter mixture-of-experts model, in preview from 6 October 2026, with open weights since 29 October under a licence that asks model-as-a-service hosts for commercial terms.
About Mistral Large 4
Mistral Large 4 is the largest model so far from Mistral AI, a mixture-of-experts system with about one trillion parameters in total and 49 billion active for each token. It reached public preview as an API on Mistral Studio on 6 October 2026.
One model for instructions, reasoning and agents
Mistral describes Large 4 as a hybrid of instruct and reasoning models with multimodal input. The aim is to handle instruction following, reasoning and agentic work in a single model rather than switching between separate ones.
Trained and served in Europe
Mistral says training used 3,800 Grace Blackwell GPUs from NVIDIA housed in its own European data centres, and that the preview is also served from them. For buyers who need their models hosted in Europe, that is the central point of the release.
Weights and open questions
Release of the weights is promised by Mistral for the end of October. In the meantime the company is testing the model adversarially with cybersecurity firms, vetted partners and government bodies. The context window and the licence for the weights are not yet stated, so anyone planning to self-host has to wait for those details. The announcement lists the preview rate per million input and output tokens, which should be read directly from Mistral before budgeting.
The company's capability claims
The post presents results in areas that include cybersecurity, finance and manufacturing. These are Mistral's own evaluations. For organisations that want a frontier-scale model with the option of running the weights themselves, the date the weights appear is the one to watch.
TrustList summary
Mistral Large 4 is the largest model Mistral AI has released: a natively multimodal mixture-of-experts model with about one trillion parameters, 52 billion of them active per token, and a context window of one million tokens. It went into public preview on Mistral Studio on 6 October 2026, and on 29 October Mistral published the weights of an improved version on Hugging Face, keeping the end-of-month date it had set.
Weights under a licence with a hosting clause
The weights come under the Mistral Ecosystem Open-Weight License, a modified MIT licence. Use, modification, fine-tuning and commercial use are allowed, and internal use is exempt from any further terms. A business that offers the model to third parties as a service, through an API or hosted endpoint that lets them control inputs, parameters or training data, must sign a separate commercial licence with Mistral before commercial use. End-user products that embed the model in specific features, and services that only relay requests to a model hosted elsewhere, fall outside that clause. The licence states no revenue threshold.
What changed after the preview
Mistral says the released model had three more weeks of reinforcement learning than the preview, and that its three reasoning modes (low, medium and high, the default) replace the preview's mode without reasoning. It also says it changed the guardrails after watching live cyber-related traffic during the preview. The model was trained from scratch on 3,800 NVIDIA Grace Blackwell GPUs in Mistral's own European data centres and adds a 1.6 billion parameter vision encoder.
Running it
The model is offered through the Mistral API and Mistral Studio, and Mistral lists OpenRouter, NVIDIA's hosted endpoints and NIM container, and Microsoft Foundry as other routes. Self-hosting is supported with vLLM and SGLang. Benchmark results in the release, such as 48% on Artificial Analysis's Coding Agent Index, are Mistral's own reporting and are not stored as scores here.
Key facts
Developer: Mistral AI, Paris Public preview: 6 October 2026 Weights released: 29 October 2026 Size: about 1 trillion parameters (1.05T per Mistral's docs), 52B active, plus a 1.6B vision encoder Context: 1 million tokens Licence: Mistral Ecosystem Open-Weight License API price: $0.70 per million input tokens, $0.07 cached, $2.10 per million output tokens
Summary written by TrustList from the business's own published information, Nov 2026.
Benchmarks & AI stats
| Benchmark | Official | Community avg |
|---|---|---|
| Artificial Analysis Intelligence Index v5.0.1llm | 26 index | - |
“Official” values are editor-approved and feed the ranking. “Community avg” is the mean of member submissions (shown for transparency; it never affects the ranking until an editor approves a value).
Sign in to add a benchmark score for this model.
Mistral Large 4 launches
Trust Score
An earned signal from verification, reviews, awards, transparency and engagement - the vendor can't buy it.
Updated 10/7/2026
What the business tells TrustList
- Founded
- 2026 (0 years) evidence sent
Request a demo or quote from Mistral Large 4
Reviews
Write the first review of Mistral Large 4
Used it? Your experience helps other buyers decide.
Write a reviewQuestions & answers
No questions yet. Be the first to ask about Mistral Large 4.
Mistral Large 4 alternatives
Similar listings buyers compare against this one.
Ai2's open-weight 8B model that writes cited scientific reports from a research question and retrieved literature
Open-source 2B decision model from AWS's Strands Agents project that picks options and returns scores with a reliability figure
Cantina Security's first open-weight security model, a post-trained GLM-5.3-Flash for vulnerability research and exploit checks
Black Forest Labs' image generation and editing model with bounding-box layouts, multiple reference images and 4K output
Open-weight translation model from Bilibili's Index team, covering 150 languages and following glossary and style instructions
Perplexity's open-weight decision model that picks among given options for routing and checks in AI systems
JetBrains' open 12B mixture-of-experts coding model with 2.5B parameters active per token under Apache 2.0
More on TrustList
Everything here links back to the same verified catalogue. Pick your next stop.
- Write a reviewHelp other buyers decide
- CompaniesAgencies, consultancies and IT service providers, ranked by verified reviews.
- ProductsSoftware and SaaS with pricing, features, integrations and alternatives.
- ArticlesGuides, comparisons and research from the editorial desk and the community.
- AwardsAnnual recognition decided by verified reviews and an independent jury.
- LaunchesNew products and releases, voted up by the community every day.
- RequestsBuyers describe what they need; vendors respond directly.
- PeopleReviewers, authors and makers with public profiles.
- ComparePut up to four listings side by side before you shortlist.