Mistral Large 4: 1T open-weight model, preview API live now
Mistral Large 4 is a 1-trillion-parameter open-weight model with 49B active. API preview is live; weights land this month. How to price it before you self-host.
Mistral released Mistral Large 4 on October 6: a 1-trillion-parameter, open-weight model that activates 49 billion parameters per token. Today it is an API preview. The weights come later this month. For a small business, "open weights" is a promise about leverage, not a server you will run, and the gap between those two is where the decision sits.
What actually happened
From Mistral's announcement and The Next Web:
- Size. About 1 trillion total parameters, mixture-of-experts, 49 billion active. It takes text and image input and supports more than 160 languages.
- Price. Mistral lists the preview at $1.36 per million input tokens and $4.18 per million output tokens on its API.
- Weights. Mistral says full weights ship by the end of October; The Next Web reports October 27. Mistral did not state the license in its announcement, so do not assume Apache 2.0.
- Hosting. The preview runs on Mistral's European data centers. Mistral says private cloud and on-premises deployment will follow.
- Claims. Mistral calls it state of the art among open models for cybersecurity, finance, and legal work. The Next Web lists 62% on DeepSWE v1.1 and 67% on FinWorkBench. Those are vendor-reported numbers.
Why it matters for your business
- You will not run this on a box in the office. A trillion parameters needs serious multi-GPU hardware even at 49B active. For most teams "open weights" means you can buy it from more than one host later, not that you self-host it.
- That is still the point. Once weights are public, other inference providers can serve it. More hosts means price competition and an exit if Mistral changes terms. That is leverage you do not get from a closed model.
- Test on your own task first. At $1.36/$4.18 it sits well under frontier closed-model pricing. Run 50 real prompts from your workflow (invoices, support replies, product copy) through the preview and compare quality and cost against what you use now.
- Read the license before you plan. Mistral's previous Large model shipped under Apache 2.0. If Large 4 lands on a custom license, check commercial-use terms before you build around it.
Key takeaways
- Mistral Large 4: ~1T parameters, 49B active, text and image input, 160+ languages
- Preview API is live at $1.36 input / $4.18 output per million tokens
- Weights ship by end of October; the license is not yet stated
- Open weights matter for multi-host pricing and exit options, not desk-side hosting
- Benchmark it on your own prompts before you switch
Want to swap models without rewriting your app? We build vendor-agnostic AI systems with an eval set from your real work, so a new model like Large 4 is a config change and a test run. See how we build or tell us what you run today.
Sources: Mistral AI, The Next Web.
- #mistral
- #open-weights
- #llm
- #self-hosting
- #model-pricing
Tommy Rush — Founder, Rush Commerce
Operator turned builder. 15+ years running operations — now shipping the systems businesses run on. More
Get The Rush Report weekly — one email, zero fluff.
Keep reading
Wajo AI agent calls businesses for customers: answer the bot
Khosla-backed Wajo's AI agent calls businesses and pays with virtual cards for its users. Here's how to get your phone line and booking flow ready.
Read itClaude gray-market API resellers: 70–90% off, but who reads it?
China's Claude 'transfer stations' resell API access at 70–90% off. Why any cheap AI API reseller is a data and model-swap risk, and how to vet yours.
Read it