🔍 Read the full analysis: Mistral Large 4: What Its Strength Outside The US And China Means For AI on ThorstenMeyerAI.com
Get the latest gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
Mistral released Large 4 as a research preview, scoring 38.4 on Artificial Analysis’s Intelligence Index. The result makes it a leading model from outside the US and China, but the source’s benchmark data puts it behind current US and Chinese flagships and shows lower-cost models outperforming it on the same index.
Mistral AI released Mistral Large 4 as a research preview, scoring 38.4 on Artificial Analysis’s Intelligence Index v4.3.2. The result makes it the highest-scoring model in the source’s comparison from outside the United States and China, but it remains below major US and Chinese competitors, complicating claims that Europe has a frontier-level alternative.
The model has 1 trillion total parameters, with 49 billion active, and supports text and image input with text output. Mistral lists a 512,000-token context window. It is currently available through the company’s API as a research preview; Mistral has said it plans to release the weights at the end of October. The source says the model’s license had not been published at the time of writing.
On Artificial Analysis’s current index, Large 4’s 38.4 score is up from 9 for Mistral Large 3 and 14 for Medium 3.5 on the same index version, according to the source. But the listed US leaders score between 52.6 and 57.6, while Chinese models including GLM-5.3, Kimi K3 and DeepSeek V4.1 Flash also rank ahead. The source characterizes Large 4 as roughly level with OpenAI’s smaller GPT-6 Luna model.
Mistral’s listed API prices are $1.36 per million input tokens and $4.18 per million output tokens, with cached input at $0.14. The source reports a 50% discount for the first two weeks. It also says Artificial Analysis measured $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; both of those models scored higher on the index. These task-cost estimates reflect the benchmark workload, not a fixed price for every customer use case.
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.
Europe’s Model Faces a Cost Test
Large 4 matters because it offers a European-developed option in a market where the most prominent frontier models come from US and Chinese labs. For governments and businesses concerned about supplier concentration or where AI infrastructure is developed, that may make Mistral’s progress relevant even if the model is not the benchmark leader.
The data also sets limits on what the launch establishes. A score of 38.4 is a substantial rise from Mistral’s earlier results cited by the source, but it does not put Large 4 alongside the top US systems. Nor does being the highest-scoring model outside the US and China, within this particular comparison, mean it leads the global field. The source’s own figures show multiple Chinese models scoring above it.
Price and performance could shape purchasing decisions as much as geography. The source’s benchmark cost estimates put two lower-priced Chinese models ahead in score, while Large 4 generated substantially more output tokens than the median comparable model. That combination may matter for agentic tasks, where repeated model calls can add expense and delay. Buyers will need to test their own workloads rather than treat one index score or cost-per-task estimate as a universal measure.
As an affiliate, we earn on qualifying purchases.
A Preview, Not a Final Release
The release is still an early-stage offering. Mistral Large 4 is available through the API as a research preview, and its weights have not yet been released. According to the source, Mistral plans to publish them at the end of October, but the license was still unpublished. That leaves open how developers will be permitted to use, adapt or distribute the model once the weights arrive.
The source attributes the benchmark comparison to Artificial Analysis Intelligence Index v4.3.2, which it says includes agentic work tests such as knowledge work, software workflows and coding. It also reports that Mistral said reinforcement learning was still running, meaning scores could change. The benchmark result should therefore be read as a measurement of the preview at this point, not a settled assessment of a final model.
Claims about hands-on reliability require a separate distinction. The source author reports seeing confident false statements during personal testing, but that is an observation rather than a published Artificial Analysis result. The source also cites hallucination figures for other models from AA-Omniscience, but those measures are not interchangeable with the Intelligence Index score and do not, by themselves, establish how a model will behave in every deployment.
“The ‘Western alternative outside the US’ race is a field of one, and Large 4 wins it by default.”
— Source article author
As an affiliate, we earn on qualifying purchases.
Weights, Licensing and Reliability
Several points remain unresolved. The planned weight-release date is at the end of October, according to the source, but the release had not happened when the material was written. The eventual license and any restrictions on commercial use are also unknown in the supplied information.
It is also unclear how much the score will move as reinforcement learning continues, or whether the final model will differ materially from the preview. The source does not provide enough detail to establish how broadly its author’s hallucination observations apply, or to compare that anecdotal testing directly with standardized reliability measures. Benchmark rankings and task-cost estimates may not predict results for a particular company’s prompts, tools or workloads.
large language model with 512k token window
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
October Weights and Updated Scores
The next stated milestone is Mistral’s planned release of Large 4’s weights at the end of October. The publication of those weights and their license will clarify whether developers can run the model independently and under what terms. Mistral may also update the preview as reinforcement learning continues; fresh benchmark results would show whether its index score and relative ranking change.
For now, prospective users can evaluate the API preview against their own requirements, including accuracy, latency, token use, cost and data-handling rules. Any comparison should separate the model’s regional significance from its measured performance: Large 4 expands Europe’s visible presence in advanced AI, while the available data does not show it matching the leading US or Chinese systems.
AI development tools for researchers
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What is Mistral Large 4?
Mistral Large 4 is Mistral AI’s multimodal model, released as a research preview through its API. The source describes it as having 1 trillion total parameters, 49 billion active parameters and a 512,000-token context window.
How does it rank against other models?
It scored 38.4 on Artificial Analysis Intelligence Index v4.3.2, according to the source. That leads the models in the source’s comparison from outside the US and China, but several US and Chinese models score higher.
Are Mistral Large 4’s weights available?
Not yet, according to the source material. Mistral planned to release the weights at the end of October, and the license had not been published when the source was written.
Is Large 4 cheaper than competing models?
Not by the source’s benchmark task-cost estimates. It reports 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. Actual costs depend on usage and workload.
Source: ThorstenMeyerAI.com
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
