🔍 Read the full analysis: How Mistral Large 4 Compares Beyond The US And China—and For Agents on ThorstenMeyerAI.com
Get business pricing on office and shipping supplies
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Mistral Large 4 scored 38.4 on Artificial Analysis’s Intelligence Index v4.3.2, a sharp improvement over earlier Mistral models but below current US frontier models and several Chinese models. The assessment also raises questions about its cost and output volume for agent tasks; Mistral says the model is still undergoing reinforcement learning, and its weights and licence have not yet been published.
Mistral Large 4, released by French AI company Mistral, scored 38.4 on the Artificial Analysis Intelligence Index v4.3.2, placing it below the leading US models and several Chinese competitors. The result marks a substantial gain for Mistral, but the independent ranking also raises practical questions for buyers considering the model for agentic work, including its relative performance, token use and cost.
Artificial Analysis’s index compares models on a mix of tasks, including agentic knowledge work, real-world work tasks, software workflows and coding. On the version cited in the source report, the highest-ranked model listed was Anthropic’s Claude Opus 5.5 at 57.6. US models occupied the top six positions, while Chinese models including GLM-5.3, Kimi K3 and DeepSeek V4.1 Flash also scored above Large 4. The ranking supports describing Large 4 as a strong model from outside the US and China, but does not put it among the index’s top performers.
The score is nevertheless a marked improvement on Mistral’s earlier results in the same index version: Large 3 scored 9, while Medium 3.5 scored 14. Those comparisons suggest a major jump for the company, though a score increase does not by itself establish how well the model will perform in every production setting. Mistral says reinforcement learning is ongoing and that scores may change.
Mistral describes Large 4 as a one-trillion-parameter model with 49 billion active parameters, text-and-image input, text output and a 512,000-token context window. It is available as a research preview through Mistral’s API. The company has said it plans to release the weights at the end of October, but the source report says the licence had not been published at the time of writing. Listed pricing is $1.36 per million input tokens and $4.18 per million output tokens, with cached input at $0.14; a 50% discount is available for the first two weeks.
Mistral Large 4: best outside the US and China — and still not a model to run your agents on
The headline is true: France has the most intelligent model outside the US and China. The independent data says the rest: every US and Chinese flagship scores higher, the best by 19 points. It costs 4× more per task than Chinese open models that outscore it, and it’s 2.5× as verbose as the median model.
~two-thirds of Opus 5.5. Level with OpenAI’s small model, Luna.
Eighth among open models once weights ship — behind seven Chinese ones. Beats GLM-5.2 and V4 Pro, loses to their successors.
Cohere doesn’t compete at this tier — reported ~14% hallucination at ~9% accuracy, because it declines most questions. A field of one.
The Index is now agentic-heavy — Briefcase, GDPval, AutomationBench, Terminal-Bench. Errors multiply across steps: tolerable in chat, fatal over a two-hour run.
AA v4.3.2Output tokens to complete the Index. On an agent, verbosity is cost and latency on every step.
AAConfident false assertions in hands-on use. US frontier has largely moved past this — Gemini 4 Argon: 15%. In fairness Chinese open models are worse (Kimi K3 51%, DeepSeek V4 Pro 94%). In an agent, a fabrication is a wrong premise every later step builds on.
AUTHOR’S TESTING · not an AA figure- Cyber defence: 50 on the AA Cyber Index; 82% CyberGym-E2E (ahead of Luna’s 78%). Likely top-3 open model on cyber.
- Documents & images: 19% GDP.pdf (+18 vs Large 3); 100 images per request.
- Speed: 116 tok/s, 1.46s TTFT — well above median.
- The jump: Large 3 scored 9 on this Index. 9 → 38 is real progress.
- Jurisdiction: French parent, EU hosting, weights promised end of October.
- Legally bound buyers (defence, classified, DORA, health data): now the best European option by a wide margin. Wait for the weights, check the licence, pilot on cyber and documents.
- Everyone else, for agentic or long tasks: don’t. A US frontier model is meaningfully more capable; GLM-5.3-Flash is more capable and 4× cheaper.
- Note: Preview — Mistral says RL is still running, so scores may move. That changes next month’s decision, not today’s.
Mistral says it has “essentially closed the gap.” It has closed the gap to where the Chinese open-weights field was a few months ago, while that field and the US frontier have both moved on. On every independent measure that matters for agents — intelligence, cost per task, verbosity and factual reliability — Large 4 is not a frontier model. “Most intelligent outside the US and China” is true mainly because almost nobody else outside those two countries is competing. Use it if you have to. Don’t use it because of the headline.
Agent Costs and Reliability
The index’s focus on multi-step work makes its result relevant beyond a general model leaderboard. Agents use models repeatedly to plan, call tools and act on intermediate results. A weaker result across those tasks may matter more over a long run than it does in a brief exchange: mistakes can carry forward, although the index score does not quantify a specific failure rate for any customer workflow.
The source report says Large 4 used 200 million output tokens to complete the index, compared with a median of 81 million for comparable models. That is a measure of output volume during this evaluation, not a universal forecast of token use. Still, where a workflow generates similarly long responses, output volume can affect both latency and cost. The report estimates a cost of $1.13 per Intelligence Index task for Large 4, compared with $0.25 for GLM-5.3-Flash and $0.27 for DeepSeek V4.1 Flash. Those two models scored higher on the same index, at 41.8 and 39.5 respectively, according to the cited figures.
For companies choosing an agent model, the figures make the trade-off concrete: a European provider may matter for sourcing or operational reasons, but geography alone does not establish that a model is the best fit. Buyers will need to test the model against their own tasks, including tool reliability, accuracy, response length and total run costs. The ranking is a comparative signal, not a substitute for that evaluation.
Popular right now
Dyson V8 Cyclone Cordless Vacuum
Shark Navigator Lift-Away Upright Vacuum (NV360)
Samsung 42-Inch S90H OLED 4K Smart TV Bundle
As an affiliate, we earn on qualifying purchases.
Mistral’s Leap in the Rankings
The source report frames the release as a significant change in Mistral’s position. Large 4’s score of 38.4 is more than four times Large 3’s score of 9 on the same index version. That is evidence of rapid progress within Mistral’s own lineup, even as the result remains below the current leaders named in the report.
The “outside the US and China” description refers to the model’s national origin, not to a top ranking across all regions or an evaluation of every AI lab. The cited comparison includes companies such as Anthropic, OpenAI and Google in the US, and Z.ai, Moonshot and DeepSeek in China. The source report says Large 4 would rank eighth among open-weight models once its weights are released, behind seven Chinese models. That is a projection based on the reported rankings, not a current open-weight ranking: the weights are not yet available.
The model’s current status also shapes comparisons. It is a research preview through an API, while its weights and licence are pending. Until they are available, prospective users cannot independently run the released weights or assess the terms under which they may be used.
“Reinforcement learning is still running, so scores may move.”
— Mistral
Open Questions on Deployment
Several practical details remain unsettled. The source report says the model’s weights are promised for the end of October and that its licence had not been published. The report does not establish whether that release schedule will hold, what licence terms will apply, or whether access and pricing will change when the weights become available.
The ranking is also a snapshot of Artificial Analysis Index v4.3.2. Mistral says training work is continuing, while the source report provides no later score. Benchmark results cannot show how Large 4 will perform in a particular company’s workflow, and the reported token total and task cost are specific to the evaluation. The report’s description of confident hallucinations comes from the author’s hands-on observations, not from an Artificial Analysis measurement; readers should treat it as an attributed observation rather than a benchmark finding.
Finally, the source material does not provide a complete account of pricing under different usage patterns, latency, service availability or performance across all languages and task types. Those factors could affect a deployment decision as much as the headline index score.
Weights, Licence and Retesting
The next stated milestone is Mistral’s planned release of Large 4’s weights at the end of October, alongside publication of the licence terms. Those details should clarify whether and how organisations can run the model outside Mistral’s API. The timing and final terms remain subject to confirmation.
Artificial Analysis scores could also change as Mistral continues reinforcement learning, according to the company. For users considering agents, the relevant next step is to compare the model with alternatives on their own multi-step workloads, recording accuracy, recovery from errors, token use and total cost. That testing would address questions the current public score cannot settle.
Key Questions
How did Mistral Large 4 score?
It scored 38.4 on Artificial Analysis Intelligence Index v4.3.2, according to the source report.
Does Large 4 lead US and Chinese models?
No. The cited ranking puts several US and Chinese models above it. The report’s geographic distinction is that it is a leading model from outside those two countries, not that it leads the full ranking.
Is Large 4 suitable for AI agents?
The index includes agentic tasks, but its score does not establish performance in every workflow. The source report raises concerns about the model’s ranking, output volume and observed hallucinations; organisations should test it on their own tasks before deployment.
Can developers download its weights now?
Not according to the source report. Mistral has said it plans to release the weights at the end of October; the model is currently offered as a research preview through the company’s API.
What does Mistral Large 4 cost?
The listed API prices are $1.36 per million input tokens, $4.18 per million output tokens and $0.14 per million cached input tokens. The source report says a 50% discount applies for the first two weeks.
Source: ThorstenMeyerAI.com
