artificialwatch
model radar / mistral / mistral-large-4

Mistral Large 4

Mistral · public preview since Oct 6, open weights since Oct 29
live · public preview 1T params · 49B active · 1M ctx $0.70 in · $2.10 out per 1M open weights
Watch this model free →Free alert about 15 min after it answers on a public API · members the second it does · no card

Large 4 went live as a public preview on Mistral's own API (Oct 6): a natively multimodal mixture-of-experts with about 1 trillion parameters, 49B active per token (52B counting embeddings and output layers, the figure in the repo name) and a 1M-token context. On Oct 29 Mistral published the open weights on Hugging Face, an improved checkpoint by its own account, and set the price at $0.70 / $2.10 per million tokens. It scores 26 on the Artificial Analysis Intelligence Index v5.0. ArtificialWatch fired the launch alert at 13:37 UTC on Oct 6, when OpenRouter listed it.

✳ the wire

Leaks & news, newest first.

Oct 29 · weights

The weights are out, and Mistral says they beat the preview

Mistral published Large 4 on Hugging Face (mistralai/Mistral-Large-4-1T-A52B, not gated, plus an NVFP4 build for Blackwell GPUs) under the Mistral Ecosystem Open-Weight License, a modified MIT that lets anyone use it commercially except a business selling it as a hosted API, which needs a separate license. Mistral says this is an improved checkpoint after three more weeks of reinforcement learning, served under the same mistral-large-4 id, and prices it at $0.70 input, $0.07 cached and $2.10 output per million tokens from today. By Mistral’s own figures Terminal-Bench 4.0 rose 13 points to 41%. Artificial Analysis’s page reads 26 on its Intelligence Index v5.0. via Mistral, Hugging Face

Oct 6 · launch

Large 4 ships as a public preview, weights by the end of October

Mistral opened a public preview API for Mistral Large 4 (id mistral-large-4): about 1.05T total parameters, 49B active, a 1.6B vision encoder and a 1M-token context, at $0.68 input, $0.07 cached and $2.09 output per million tokens for now (Mistral’s docs strike through a $1.36 / $0.14 / $4.18 list price). By Mistral’s own figures it scores 61.7% on DeepSWE v1.1 and 28.3% on Terminal-Bench 4. Trained from scratch on 3,800 Grace Blackwell GPUs in Mistral’s European datacenters. Weights were promised by the end of the month (they landed Oct 29). via Mistral

Sep 23 · status

Summer ends without the promised model

Mistral’s docs for the season list OCR 4.1 reaching general availability and Z.ai’s GLM-5.3 added as a hosted third-party model (Sep 15), but no new general-purpose Mistral model — neither the summer release Mensch described nor a Large 4. Large 3 is still the top of the line. via Mistral docs

Jul 7 · status

Still nothing — and the gap is the story

No Large 4 teaser, preview or roadmap entry. Seven months since a flagship release is uncharted for Mistral, and prediction markets read the silence as a second-half story.

Jul 4 · on record

Mensch: an open-weight model “this summer,” early access in July

Mistral CEO Arthur Mensch told TechCrunch: “We have a very exciting model to come this summer — it will be open-weight, and we’re opening early access to it in July.” He did not name it or say which line it belongs to. via TechCrunch

Mar 16 · launch

Small 4 ships — three products folded into one

Mistral merged Magistral (reasoning), Pixtral (vision) and Devstral (agentic coding) into Small 4 — a consolidation play that reads like the platform being tidied ahead of a bigger swing. via Mistral changelog

Mar · context

The March spree: six products in one month

Small 4, Voxtral TTS, Leanstral, Forge, the Spaces CLI and an NVIDIA partnership — everything except a flagship. The Large line stayed untouched. via Serenities AI

Dec 2 · flagship

Large 3 ships — the current top of the line

Mistral's reigning flagship arrived December 2, 2025. Every day since is a day closer to its successor. via Mistral

✳ from the wire desk

The official word.

“Today, we’re launching a public preview of Mistral Large 4.” Mistral said the weights would follow by the end of the month (they landed Oct 29), and that until then cybersecurity partners and state authorities are red-teaming a version with reduced moderation. Our launch alert fired at 13:37 UTC, when OpenRouter listed it.

Mistral, Oct 6, 2026 · via the ArtificialWatch wire desk

Large 3The previous flagship, shipped Dec 2, 2025.
Small 4March's consolidation: reasoning + vision + agentic coding in one.
Large 4Live since Oct 6 (public preview); open weights since Oct 29.
✳ pricing

$0.70 / $2.10 per million tokens, on Mistral's own API.

How much does Mistral Large 4 cost?

Mistral's own API lists Mistral Large 4 (mistral-large-4) at $0.70 per million input tokens and $2.10 per million output tokens, and $0.07 per million cached input tokens. Prices as published in the models.dev catalog, as of 2026-11-21.

What is Mistral Large 4's context window?

1.05M tokens, with up to 262K output tokens per response, as Mistral lists Mistral Large 4 (mistral-large-4).

✳ where to use it

Listed by 16 providers, exact ids below.

16 providers list Mistral Large 4 in the models.dev catalog (as of 2026-11-21), under the model ids shown below.

Mistralmistral-large-4
OpenRoutermistralai/mistral-large-4-0
Vercel AI Gatewaymistral/mistral-large-4
CloudCode.ONEmistral-large-4
CommandCodemistral/mistral-large-4
Cortecsmistral-large-4-0
DevPass (LLM Gateway)mistral-large-4
Eden AImistral/mistral-large-4@eu
Kilo Gatewaymistralai/mistral-large-4-0
LLM Gatewaymistral/mistral-large-4
LowRouterauto/mistralai/mistral-large-4
NanoGPTmistralai/mistral-large-4

Also listed by Ollama Cloud, OpenCode, Requesty, Standard Compute.

Listings cover the model family, so a provider may list a sibling tier under the same name. Check the id before you build on it.

Head to head, on the public leaderboards:

Large 4 landed Oct 6. Hear the next frontier launch first.

Mistral Large 4 went to public preview on Oct 6; our launch alert fired at 13:37 UTC, when OpenRouter listed it. We keep sweeping the public catalogs (OpenRouter, models.dev) every 60 seconds. When the next frontier model returns a real answer, a text or a phone call fires on paid tiers; free email + Chrome push follow about 15 minutes later.

On the board. You'll hear the second the next frontier model answers.