🔍 Read the full analysis: Why Mistral Large 4 Is Worth A Look Beyond The US And China on ThorstenMeyerAI.com
Get hardware and tech essentials delivered free — and shop member deals
- Fast, free delivery on millions of items
- Access to Prime Big Deal Days deals on October 6–7
- Prime Video, Amazon Music and more included
TL;DR
Mistral Large 4 scored 38.4 on Artificial Analysis’s Intelligence Index, making it a leading model from outside the US and China but leaving it below current US and Chinese flagships in the supplied comparison. Its progress from earlier Mistral models is substantial, yet benchmark performance, per-task costs and still-unpublished weight terms leave its practical value unsettled.
French AI company Mistral released Large 4 in a research public preview, and Artificial Analysis’s Intelligence Index v4.3.2 gives it a score of 38.4. The result makes it a strong contender among models from outside the United States and China, but the cited benchmark puts it below the leading US and Chinese models—and its reported price per benchmark task is higher than that of two Chinese models that score above it.
Large 4 is a one-trillion-parameter model with 49 billion active parameters, according to the source. It accepts text and images, produces text, and supports a 512,000-token context window. Mistral has made it available through its API as a research public preview. The company says it plans to release the model weights at the end of October; until then, the model is proprietary and its licence has not been published, the source reports.
On the cited Artificial Analysis index, Large 4 scored 38.4, compared with 9 for Mistral Large 3 and 14 for Medium 3.5 on the same index version. That marks a sharp improvement for Mistral. But the supplied comparison lists US models at the top, including Anthropic’s Claude Opus 5.5 at 57.6 and OpenAI’s GPT-6 Astra at 52.7. Chinese models also score above Large 4 in the table, including GLM-5.3 at 44.8 and DeepSeek V4.1 Flash at 39.5.
The source gives API prices of $1.36 per million input tokens and $4.18 per million output tokens, with cached input at $0.14 per million tokens. It says Mistral offered a 50% discount for the first two weeks. Artificial Analysis’s task-cost comparison puts Large 4 at $1.13 per benchmark task, against $0.25 for GLM-5.3-Flash and $0.27 for DeepSeek V4.1 Flash. Those figures apply to the benchmark’s tasks; they do not establish what every customer will pay for a different 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.
Performance Meets Procurement Costs
The result matters because buyers evaluating AI models need to weigh capability, cost and operational fit, not just a model’s national origin or its headline benchmark position. Large 4’s move from earlier Mistral scores suggests the company has made substantial progress. At the same time, the comparison supplied here does not place it at the frontier of the tested models, and the reported task costs complicate the case for choosing it on value alone.
That trade-off may be especially relevant for multi-step agent workflows, where a model must sustain a sequence of decisions rather than answer a single prompt. The source reports that Large 4 generated 200 million output tokens across the Intelligence Index tasks, compared with a median of 81 million for comparable models. If those measurements apply to a buyer’s workload, greater output volume can add cost and time. Benchmark results, however, cannot by themselves predict performance in every company’s systems.
The source author also reports seeing confident false statements in hands-on testing. That is an attributed observation, not a finding from the cited index. It is a reason for prospective users to test reliability and verification requirements in their own environment, rather than assume that a high-level benchmark score answers those questions.
As an affiliate, we earn on qualifying purchases.
A Fast Jump From Mistral 3
The supplied account compares Large 4 with earlier Mistral models using the same version of Artificial Analysis’s index: Large 3 scored 9 and Medium 3.5 scored 14, while Large 4 scored 38.4. That makes the release a major step up within Mistral’s own lineup. It does not erase the gap shown in the wider table, where the highest-listed US model scores 57.6 and several Chinese models also exceed Large 4.
The index described in the source draws heavily on agentic and work-oriented evaluations, including AA-Briefcase, GDPval-AA, AutomationBench and Terminal-Bench 4.0. Its score is therefore a measure across those listed tests, not a complete assessment of every use case, such as simple chat, image understanding or a specific company’s retrieval system. The source also says Mistral’s reinforcement-learning work is ongoing and that scores may change.
The article’s claim that Large 4 is the most intelligent model outside the US and China should be read in that limited frame. It reflects the rankings and comparisons supplied by the source, not proof that the model is best for every task or that no other developer could compete under a different test.
As an affiliate, we earn on qualifying purchases.
Preview Terms and Reliability
Several practical details remain unsettled in the supplied material. The weights are not yet available, and the model’s licence has not been published. The source says Mistral plans to release the weights at the end of October, but gives no further detail on access conditions or whether that schedule will hold.
The reported benchmark and pricing data also do not settle how Large 4 will perform or cost in a particular deployment. Actual expenses depend on a user’s token volumes, prompt and output patterns, and the applicable price at the time. Mistral’s ongoing reinforcement learning may alter benchmark results, according to the company as quoted in the source.
Finally, the source author’s account of hallucinations comes from hands-on testing, without details here about the number of tests or the testing protocol. Artificial Analysis’s index score is separate from that observation. The material provided does not establish a broad, independently measured hallucination rate for Large 4.
As an affiliate, we earn on qualifying purchases.
Weights and Further Testing
The next stated milestone is Mistral’s planned release of Large 4’s weights at the end of October. Their publication, alongside licence terms, should make it clearer whether developers can run or adapt the model outside Mistral’s API and under what conditions. The timing and terms remain subject to confirmation.
In the meantime, Mistral’s research preview is available through its API, and the company says reinforcement learning is continuing. Buyers considering the model can compare updated benchmark figures and test their own workloads, including output volume, accuracy and performance over multi-step tasks. The supplied source does not identify a later benchmark date or a confirmed production release schedule.
text and image AI processing tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What is Mistral Large 4?
It is a Mistral model available in a research public preview through the company’s API. The source describes it as a one-trillion-parameter, natively multimodal text-and-image input model with a 512,000-token context window.
How does Large 4 compare on the cited benchmark?
Artificial Analysis’s Intelligence Index v4.3.2 scores it at 38.4. The supplied table lists several US and Chinese models above it, including Claude Opus 5.5 at 57.6 and GLM-5.3 at 44.8.
Is Large 4 open-weight now?
No. The source says it is currently a proprietary preview. Mistral plans to release the weights at the end of October, but the licence has not been published in the supplied material.
How much does the API cost?
The source lists prices of $1.36 per million input tokens, $4.18 per million output tokens and $0.14 per million cached input tokens. It also reports a 50% discount for the first two weeks; customers should check current pricing because the discount period and rates may change.
Do the benchmark results prove Large 4 is unreliable?
No. The source author reports seeing confident false statements during hands-on testing, but the provided account does not include a test protocol or a broad, independently measured hallucination rate for Large 4. Buyers should treat that report as an attributed observation and evaluate the model on their own tasks.
Source: ThorstenMeyerAI.com
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
