On the Artificial Analysis ranking consulted on 28 October 2026, the top model with downloadable weights scores 32 points, versus 58 for Anthropic’s Claude Opus 5.5. With version 5 of the index, which aggregates multiple tests, open-weight models all lose between 12 and 16 points, while the models themselves have not changed.
- Le nouvel indice Artificial Analysis fait perdre 12 à 16 points aux modèles ouverts.
- Ce nouvel indice repose à 60 % sur des tests privés.
- Mistral Large 4 passe de 38 à 25 points, et ses poids ne sont pas encore publiés.
- L’écart avec les modèles fermés atteint 26 points si l’on ne compte que les poids réellement publiés.
In the previous version, still online a few days earlier, Xiaomi’s MiMo-V2.6-Pro, DeepSeek V4.1 Flash and GLM-5.3 all exceeded 44 points. The new version indeed replaces public evaluations with question sets that laboratories cannot consult.
60 % of the index now relies on private tests
According to the presentation article for version 5, 60 % of the index is based on evaluations whose test sets are private or whose answers are kept secret. The index is divided into four families, each weighting 25 %: intellectual work, code, science and general knowledge.
Artificial Analysis notably cites FrontierCode 2.0, a programming benchmark launched by Cognition, whose tasks “remain private to avoid contamination.” Contamination refers to the presence of test questions in the models’ training data. The index also includes 40 hard physics questions from CritPt v2, while AA-Omniscience v1.1 has been stripped of its public questions. The article does not detail the exact composition of the 60 %.
Three tests drop out of the index: GDPval-AA, SciCode and AA-LCR. Two are added, FrontierCode 2.0 and Terminal-Bench-Science, dedicated to scientific tasks performed in a terminal. On its X account, Artificial Analysis had already raised the private share to 40 % with version 4.2 on 4 September, and announced an increase in version 5.
Open models lose between 12 and 16 points
The eleven models listed on the open-models page, still displayed in version 4.3.2 on 28 October, all lose between 12 and 16 points in version 5. Mistral Large 4, whose weights are not yet released, loses 13. MiMo-V2.6-Pro thus drops from 46 to 30, DeepSeek V4.1 Flash 1017 (max) from 45 to 32, and Muse Spark Open (Meta, xhigh) from 39 to 25.
Moonshot AI’s Kimi K3.1 Pro, whose weights are not released, falls from 55 to 46. The top-ranked closed models remain between 50 and 58 points: Claude Opus 5.5 (58), GPT-6 Astra (57), GPT-6.1 Sol (55), Claude Sonnet 5.5 (54), GPT-6.1 Terra (51) and Gemini 4 Argon (50). Artificial Analysis does not explain the drop of the open models.
From 3 to 26 points gap depending on the model counted as open
Artificial Analysis summarizes version 5 by writing that the gap for open-weight models “drops from 3 to 12 points.” This calculation includes Kimi K3.1 Pro at 46 points, even though its weights are “not yet released.” In its 23 October article, the site announced that the gap would shrink “from 12 points to 3” once those weights were released, but they were still not released on 28 October. The model’s sheet even classifies it as “Proprietary model.”
Based on the only weights actually released, DeepSeek V4.1 Flash and GLM-5.3, at 32 points, are 26 points behind Claude Opus 5.5. Before the version change, the gap was 12 points. With 25 points, Mistral Large 4 joins Muse Spark Open, and DeepSeek V4.1 Flash and GLM-5.3 lead these two models by 7 points.
The ranking also shows each model’s reasoning‑effort setting. Set to low effort, Claude Opus 5.5 scores 35 points for an average cost of $0.98 per index task, according to Artificial Analysis. At maximum effort, the same task costs $10.04.
The Hugging Face repository for Mistral Large 4 lists 31 October 2026 as the estimated release date for its weights, with no file or license as of the evening of 28 October. Mistral wrote on 6 October that it would publish them “by the end of the month.” The chosen license and the score of the released version remain unknown.
