🔍 Read the full analysis: Why Mistral Large 4 Has Appeal Outside The US And China But Not 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 has released Large 4 as a research preview, presenting it as a leading model from outside the United States and China. Artificial Analysis scores it at 38.4, below major US and Chinese models; the source also reports high task costs and verbosity, while concerns about hallucinations come from hands-on observation rather than the benchmark.
Mistral released Large 4 as a research preview, positioning the French model as a leading option outside the United States and China. But the Artificial Analysis Intelligence Index v4.3.2 gives it a score of 38.4, behind the major US frontier models and several Chinese systems—an important distinction for buyers considering it for agent-based work.
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 the company’s API as a research preview; Mistral says it plans to release the weights at the end of October. Until then, the source says, the model is proprietary and its licence has not been published.
Artificial Analysis’s index gives Large 4 a score of 38.4. That is below listed scores for US models including Claude Opus 5.5 at 57.6 and Gemini 4 Argon at 52.6, as well as Chinese models such as GLM-5.3 at 44.8 and DeepSeek V4.1 Flash at 39.5. The source describes Large 4 as the strongest model from outside the US and China, but the benchmark table also shows it trailing several Chinese models.
The source reports standard API prices of $1.36 per million input tokens and $4.18 per million output tokens, with cached input priced at $0.14. It says Mistral is offering a 50% discount for the first two weeks. The source’s benchmark-based estimate puts Large 4 at $1.13 per Intelligence Index task, compared with $0.25 for GLM-5.3-Flash and $0.27 for DeepSeek V4.1 Flash. Those figures are source-reported comparisons, not a guarantee of costs for a particular customer’s workload.
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.
Why Agent Workloads Raise Concerns
The main question is not simply whether Large 4 can answer prompts; it is whether its performance and cost suit multi-step agent workflows, where a model must make a sequence of decisions and use tools. A weakness at one step can affect later steps, so a lower benchmark score may matter more for a long-running task than for a short exchange. The index includes agent-oriented evaluations, according to the source, but a score alone does not establish how a particular deployment will perform.
Cost can also accumulate across repeated calls. The source says Large 4 generated 200 million output tokens while completing the index, compared with a median of 81 million for comparable models. That is a benchmark observation, not a universal measure of verbosity in every use case. Still, if a workflow produces similarly long responses, output-token charges and latency could make it harder to justify than a lower-cost model with a higher score.
The source author also reports seeing confident false statements in hands-on testing. That observation is not an Artificial Analysis finding and is not quantified in the supplied material. It remains relevant as a caution for agent design: an incorrect statement can become a premise for later actions, so systems handling consequential work need suitable checks and human review.
Products Worth Considering
FLEXISPOT 60 × 24-Inch Electric Standing Desk, White
Apple MacBook Neo 13-inch Laptop with A18 Pro Chip, 8GB Memory, 512GB SSD, Liquid Retina Display, and Touch ID (Blush)
Samsung 34-Inch Odyssey G5 Ultra-Wide Curved Gaming Monitor
As an affiliate, we earn on qualifying purchases.
A Sharp Jump From Large 3
The same index version scored Mistral Large 3 at 9 and Medium 3.5 at 14, according to the source. Large 4’s 38.4 therefore marks a substantial improvement on that comparison. The score supports a claim of progress by Mistral, but it does not place the model alongside the highest-ranked US systems or above the Chinese models listed in the table.
Mistral says reinforcement learning is still underway and that scores may change. The result is consequently a snapshot of a research preview, not necessarily the final performance of the model after further work. Comparisons also depend on the benchmark, the models selected and the version evaluated; the figures here refer to Artificial Analysis Intelligence Index v4.3.2 as reported by the source.
“Reinforcement learning is still running, so scores may move.”
— Mistral
Preview Scores and Licensing
Several details remain open. Mistral has not yet released the model weights, and the source says the licence is unpublished; it is not clear from the supplied material what access or usage terms will apply after release. The date is given as the end of October, without a year specified in the source.
Performance may shift because Mistral says training work is continuing. The source does not provide enough information to establish whether the reported token volume or task costs will match typical customer workloads. Nor does it provide a systematic, independently verified hallucination rate for Large 4. The author’s testing observation should not be treated as a measured rate or a settled comparison with other models.
Weights and Further Evaluation
The next stated milestone is Mistral’s planned release of Large 4’s weights at the end of October. That release should clarify the licence and make it possible to evaluate the model outside its current API preview, if the announced schedule holds. Mistral has also said reinforcement learning continues, so updated evaluations may alter the current benchmark picture.
For organisations weighing deployment, the immediate next step is to test the model against their own tasks, measuring accuracy, tool use, response length, latency and total cost. The current comparison suggests caution for long-running agents, but the supplied figures do not settle how Large 4 will perform in every workflow or how its capabilities may change.
Key Questions
What is Mistral Large 4?
Mistral Large 4 is a multimodal model released as a research preview through Mistral’s API. 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’s index?
It scored 38.4 on Artificial Analysis Intelligence Index v4.3.2, according to the source. The listed results put it below several US and Chinese models.
Why does the source question using it for agents?
The source points to its benchmark position, reported output-token use and estimated cost per task. Its author also describes seeing confident errors in hands-on testing, but that observation is not a quantified benchmark result.
Can developers download Large 4’s weights now?
Not according to the supplied material. The model is currently a proprietary API preview, and Mistral plans to release the weights at the end of October. The licence has not yet been published.
Source: ThorstenMeyerAI.com
