Mistral Large 4
Mistral's 1T preview scores no higher than 72 in any job; weights due 31 Oct. Use GLM-5.3 or Kimi K3 now.
Should you use it?
Pick it if
Teams that want a European open-weight model and can wait for the weights and final scores.
Skip it if
You need strong coding or agent results today; MiMo-V2.6-Pro is open-weight, downloadable now, scores 52 for coding against 36 here, and ranks 13th on Arena's Agent board against 39th.
The short version
- What this is
- Mistral's new flagship model, a 1-trillion-parameter open-weight model that reads text and images, in public preview on Mistral's API since 6 Oct 2026.
- Why it matters
- Mistral calls it the most powerful open-weight model outside China s2; until the weights ship and scores settle, it is news, not a pick.
- Public preview on Mistral's API since 6 Oct; weights due 31 Oct per Mistral's Hugging Face page s1 s8.
- $0.68 / $2.09 per 1M tokens during the launch sale; list $1.36 / $4.18 s1.
- Scores 36 for coding, 72 for research and 59 for writing; 39th on Arena's Agent board s9 s6 s10 s11 s12.
Good at
Weak at
- Weights are not out yet; Mistral's Hugging Face page gives a current ETA of 31 October 2026 s8.
- Mid-table or lower on arena.ai: WebDev 42 of 142, Expert 108 of 364, Agent 43 of 52, Creative Writing 157 of 412 s9 s10 s6 s12.
- Every benchmark score so far is Mistral's own and preliminary; reinforcement learning is still underway s2 s3.
Our take
Large 4 is Mistral's biggest model, and the launch got wide coverage s2 s3. You can call it on Mistral's API now as a public preview, at a launch-sale price of $0.68 / $2.09 per 1M tokens s1. Mistral's changelog says "Open weights are coming soon" s7, and its Hugging Face page gives a current ETA of 31 October s8. The sale price is half the list price of $1.36 / $4.18 s1.
The public numbers are not strong yet. It scores 36 for coding, from rank 42 of 142 on arena.ai WebDev s9, and ranks 43 of 52 on the Agent board s6. Research and writing score 72 and 59 s10 s12. Mistral says training is still going on s3, so we'll score it again when the weights ship. For an open model today, Kimi K3 and GLM-5.3 rank higher on WebDev s9.
Scores by job
A score from 0 to 100 per job, against everything available today. Overall 52 across 3 jobs, the average of the job scores.
New to this? What the scores and labels mean
- 0 to 100
- Our score for one job, like writing or coding, compared with everything else out today. 100 is the best available today. For AI models we build it from public data, such as Arena's blind votes and Epoch AI's independent tests; apps and tools are scored on our own checklist. The same tool can score differently for different jobs, and where two likely ranges overlap, we call it a tie.
- Our verdict
- Use it: a top choice for most people. Worth trying: good, and worth a go. Depends: good for some people or one kind of job. Wait: too early to tell, or not out for everyone yet. Skip: something else does the job better.
- First look
- Just released. We say what it is and whether to bother, and give it a score once public results exist.
- No score yet
- None of the public sources we use covers it yet, or the only evidence is from a source whose numbers we can't use, so we describe it in words instead.
- Best buy
- Nothing cheaper scores as well for that job.
- Best free
- The best score for that job that costs nothing, through a free plan or by running on your own computer.
- Pay as you go
- No monthly plan: you pay for each use, usually through a developer account.
- Checked
- The day we last confirmed the score and the price against the sources. Prices change often, so check the date.
- AI model
- The engine underneath, like GPT or Claude. You use it through an app or, if you code, directly.
- Open source
- Free to download and run on your own computer, so nothing leaves your machine.
#13 of the 15 we have scored for this job.
From blind votes (Arena) and independent tests (Epoch AI) · Arena data, CC BY 4.0, Epoch AI data, CC BY 4.0
2 sourcesData read 20 Oct
#11 of the 12 we have scored for this job.
From blind votes (Arena) · Arena data, CC BY 4.0
1 source so farData read 20 Oct
#21 of the 21 we have scored for this job.
From blind votes (Arena) · Arena data, CC BY 4.0
1 source so farData read 20 Oct
#39 in Arena's blind votes for agents. Those votes give a rank but no rating, so there is no score yet.
#39 in Arena's blind votes for agents (rank only, no score yet) · Arena data, CC BY 4.0
2 sourcesData read 20 Oct
Alternatives
Better or cheaper for agents, from our scores, then the ones we'd also look at. The same kind of product comes first.
- Claude Opus 5.5BetterText model#1$20 a monthRanked #1 on Arena's Agent board in agents, against #39 here; $20 a month. Side by side
- MiMo-V2.6-ProFreeText model#13FreeRanked #13 on Arena's Agent board in agents, with a free plan. Top open-weights model on the Artificial Analysis index today. Side by side
- GLM-5.3Text modelNot reviewed yetAn open-weight model already on the public boards, with weights you can download now.
- Kimi K3Text modelNot reviewed yetOpen model that ranks well on arena.ai Text today.
Value for money
Where it sits in agents: higher is a better score, further left is cheaper to run. Green dots are the best value.
Evidence
Everything behind the call and the scores, folded so the page stays short. Spotted a wrong price, score or link? Report a mistake.
How we scored this
The research, writing and coding scores are our own, built from public data: blind votes on Arena and, where they cover the job, Epoch AI's independent tests. How we combine them is our method and we don't publish it. For agents, Arena's blind votes give a rank but no rating, so we show the rank and no score yet.
Public leaderboardIndependent test
Reviews are drafted with help from AI models, including Anthropic's Claude. Model scores come only from outside leaderboards and independent evaluations, by a written rule.
No one paid for this review or gave us free access, and we earn nothing if you sign up. How we score
Sources 13
- s1Mistral docs: Mistral Large 4 model card
- s2The Next Web: Europe's Mistral launches Large 4
- s3BetaNews: Mistral opens trillion-parameter Large 4 preview
- s4Mistral Help Center: opting out of data use for training
- s5arena.ai Text: Coding leaderboard
- s6arena.ai Agent leaderboard
- s7Mistral changelog
- s8Hugging Face: mistralai/Mistral-Large-4-1T-A52B
- s9arena.ai Code Arena: WebDev leaderboard
- s10arena.ai Text: Expert leaderboard
- s11Artificial Analysis LLM leaderboard
- s12arena.ai Text: Creative Writing leaderboard
- s13arena.ai Text: Instruction Following leaderboard
Something here out of date or wrong? Report a mistake and we will check it against the source.
History 4 changes
Every change to a rating, the price or our call, with the reason. Scores are against what exists on the day, so they drift down as better things ship. Older entries come from before we moved to scores.
- First look: public preview, weights not yet released, too few votes to rating.
- Price re-confirmed on Mistral's model page: list $1.36 / $4.18 per 1M tokens, on sale at $0.68 / $2.09; 1.05T total, 52B active, 1M context. Weights ETA now 31 Oct 2026 from Mistral's Hugging Face page, replacing the 27 Oct press date.
- Audit: added live arena.ai WebDev rank 42 of 142 and Text Expert rank 108 of 364; still waiting on the weights.
- New: First look to ScoredNew: Coding ratedNew: Agents ratedNew: Research ratedNew: Writing ratedRated: it is publicly available as an API preview and its arena.ai rows are not marked preliminary. Coding and agents scored highest; research (moved up a step by the Artificial Analysis index) and writing (moved up a step by Instruction Following) scored lower. Verdict stays wait: weights are due 31 Oct and Mistral says training continues.
Use it well 4 guides on this site
Pages on this site that help with the jobs above.