Mistral Large 4 has scored 38 on Artificial Analysis’s Intelligence Index, matching OpenAI’s GPT-6 Luna (max) and trailing DeepSeek V4.1 Flash by one point. The same comparison puts five Chinese open models ahead, with Xiaomi’s MiMo-V2.6-Pro at 46, while Large 4 posts a stronger result on one cybersecurity test.
Mistral AI Watch analysis
What happened
Artificial Analysis ranks Large 4 as the highest-scoring model from outside the US and China in its Intelligence Index, according to a report published by lavx.hu. The model scored 38; MiMo-V2.6-Pro scored 46, and the report lists four other Chinese models above Large 4 as well.
The comparison is not a clean victory on cost. Artificial Analysis estimates Large 4 at $1.13 per task, or $0.57 during a reported launch discount, against $0.13 for MiMo-V2.6-Pro. Large 4 did better on CyberGym-E2E-AA, scoring 81.7% against 79% for MiMo-V2.6-Pro and 78% for GPT-6 Luna (max). But its overall Cyber Index score was 50, level with GLM-5.3-Flash and below two models at 56.
Why it matters
The results make Large 4 more interesting than a single leaderboard number suggests. It appears competitive on broad intelligence and particularly strong on that cybersecurity test, but the costs and other cyber results tell a less triumphant story. The report says Artificial Analysis expects it to rank among the top three open-weight models on the Cyber Index once its weights are released; that is a projection, not a completed result.
For teams comparing models, the practical lesson is to match the benchmark to the job. A high score on one cybersecurity test does not settle the wider cyber picture, and a model’s task cost can matter more than its headline rank.
Our read
Large 4 looks like a credible European contender, not a leaderboard coronation. Its most persuasive case is the mix of a strong intelligence score and a standout result on one cyber test. The awkward footnote is that several Chinese models combine higher index scores with much lower reported task costs. If you are choosing a model, test the specific workload and wait for the promised weights before treating these comparisons as the final word.
What to watch
- Whether Mistral releases the model weights by the end of October, as the report says it plans to.
- How Artificial Analysis scores Large 4 once those weights are available.
- Whether independent tests reproduce the cybersecurity result across a wider range of tasks.
- Whether Mistral’s prices or performance change as post-training continues.
Discussion spark: For a real deployment, would you favour Large 4’s reported cybersecurity result, or a cheaper model that scores higher overall?
Sources and evidence
not affiliated with or endorsed by Mistral AI