🔍 Read the full analysis: Mistral Large 4: Why Its Global Appeal Doesn’t Make It Right For Agents on ThorstenMeyerAI.com
Get business pricing on tech for your team
- 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’ Intelligence Index v4.3.2, a sharp rise from Mistral Large 3’s 9 on the same index version. The supplied analysis says it remains behind leading US and Chinese models and costs more per benchmark task than two Chinese models that score higher; the model is still a research preview, with weights and licensing details pending.
Mistral released Large 4 as a Research Public Preview, and the model scored 38.4 on Artificial Analysis Intelligence Index v4.3.2, according to the supplied analysis. That is a major improvement over Mistral’s earlier scores, but the same comparison places Large 4 below leading US and Chinese models and identifies cost, output volume and reported hallucinations as concerns for customers considering it for agent workflows.
Large 4 has 1 trillion total parameters, with 49 billion active, and accepts text and images while producing text. Mistral lists a 512,000-token context window. The model is available through Mistral’s API as a preview; the source says Mistral has promised to release weights at the end of October. Until then, the model is proprietary, and its licence has not been published.
Artificial Analysis’ v4.3.2 results put Large 4 at 38.4 points. The supplied comparison lists Claude Opus 5.5 at 57.6, GPT-6 Astra at 52.7 and Google’s Gemini 4 Argon at 52.6. Chinese models in the table also score higher, including GLM-5.3 at 44.8, Kimi K3 at 43.6 and DeepSeek V4.1 Flash at 39.5. The figures are benchmark results, not a guarantee of performance on every customer’s tasks.
The analysis reports standard API prices of $1.36 per million input tokens and $4.18 per million output tokens, with cached input at $0.14. It says Mistral is offering a 50% discount for the first two weeks. On Artificial Analysis Index tasks, the source reports a cost of $1.13 per task for Large 4, compared with $0.25 for GLM-5.3-Flash, which scored 41.8, and $0.27 for DeepSeek V4.1 Flash, which scored 39.5. Those are source-reported benchmark costs; actual bills will depend on usage and task design.
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.
The Cost of Running Agents
The results matter because the index used in the supplied analysis includes agentic work benchmarks, such as knowledge work, SaaS workflows and coding tasks. An agent may take many model calls to complete a job. If performance falls short at each step, errors can accumulate; a confident but false answer may also become an assumption that later steps rely on.
The source analysis says Large 4 generated 200 million output tokens across index tasks, compared with a median of 81 million for comparable models. That benchmark observation suggests greater output volume in those tests, but does not establish that every real-world use will be equally verbose. If the pattern carries over, output tokens add both cost and latency across multi-step runs.
These trade-offs affect procurement decisions. A lower benchmark score alone does not rule out a model for a particular workflow, and the model could suit tasks where its multimodal input or other characteristics are useful. But buyers comparing agents need to test task success, error recovery, token use and total cost—not just headline capability or the price per token.
As an affiliate, we earn on qualifying purchases.
A Sharp Rise From Large 3
The source reports that Mistral Large 3 scored 9 and Medium 3.5 scored 14 on the same Artificial Analysis index version. Large 4’s 38.4 therefore marks a steep improvement in one release. That is meaningful progress for Mistral, even though the result does not put its flagship alongside the highest-scoring models in the supplied table.
The headline that France has the most intelligent model outside the United States and China is attributed in the source to Artificial Analysis. It is a geographic comparison, not a claim that Large 4 leads the global field. The supplied analysis argues that the distinction has limited purchasing value when models from the US and China are the principal competitors in its benchmark comparison.
Mistral says reinforcement learning is still underway, so scores may change. The current results describe a preview model and a particular index version; they should not be treated as a final assessment of future releases or every deployment.
large language model API subscription
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Preview Results Still Have Limits
Large 4 is not yet a released-weights model, according to the source, and its licence has not been published. The promised end-of-October weights release and the terms under which they may be used remain pending. The source also says reinforcement learning is ongoing, meaning current benchmark scores may not represent the eventual model.
The source author reports seeing confident hallucinations in hands-on use. That is an attributed observation, not an Artificial Analysis benchmark finding, and the material does not give a testing protocol, sample size or reproducible results. It should be weighed differently from the index scores. The supplied material also does not establish how the model performs across customer deployments or whether later tuning will affect error rates, verbosity or cost.
Benchmark task costs and token totals are tied to Artificial Analysis evaluations. They do not by themselves predict costs for a company’s workload, which depends on prompts, task length, retries and model settings. The source material ends before completing its comparison with Gemini 4 Argon, so it does not provide a full conclusion about that model’s relative cost-value trade-off.
As an affiliate, we earn on qualifying purchases.
Weights and More Testing Ahead
The next stated milestone is Mistral’s planned weights release at the end of October. Buyers will need to see the accompanying licence before judging whether the model can be deployed or adapted under terms that fit their requirements. Mistral’s API preview and continuing reinforcement-learning work provide the current route to test the model, but the source does not give a date for a final release.
For teams evaluating agent use, the practical next step is to run Large 4 against their own tasks and compare it with alternatives on completion rates, factual errors, recovery from failed steps, latency and total token costs. Future Artificial Analysis results may also clarify whether the current ranking changes as training continues. No later score or product update is confirmed in the supplied material.
As an affiliate, we earn on qualifying purchases.
Key Questions
What is Mistral Large 4?
Mistral Large 4 is a multimodal model available through Mistral’s API as a Research Public Preview. The source describes it as having 1 trillion total parameters, 49 billion active parameters and a 512,000-token context window.
How did Large 4 score on Artificial Analysis?
It scored 38.4 on Intelligence Index v4.3.2, according to the supplied analysis. The same source reports scores of 9 for Mistral Large 3 and 14 for Medium 3.5 on that index version.
Why does the source question using it for agents?
The analysis points to a gap from higher-scoring models, higher reported task costs than two Chinese models that score above it, and greater output volume in the index tests. Its author also reports observing confident hallucinations in hands-on use; that observation is not an index finding.
Can customers download or use the weights?
The source says Large 4 is currently available through Mistral’s API as a preview, with weights promised for the end of October. It also says the licence has not been published, so download availability and usage terms remain unconfirmed in the supplied material.
Source: ThorstenMeyerAI.com
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
