- Artificial Analysis released version 5 of its model intelligence ranking; about 60% of the tests are now private.
- In the new ranking, the gap between the best closed and open models grew from 3 to 12 points.
- Cost per task among the top 10 models differs by more than 13 times.
In short: on 27 October 2026 Artificial Analysis released version 5 of its AI model intelligence ranking. It changed the tests, and about 60% of the score now comes from private test sets. According to the ranking’s authors, the gap between closed and open models grew from 3 to 12 points.
What changed
- About 60% of the ranking rests on private test sets or held-out answers; that is the largest share since version 4 launched in January.
- Two new tests were added (coding and science tasks in a terminal, 15% each) and three old ones were removed.
- Claude Opus 5.5 leads with 58, followed by GPT-6 Astra (57) and GPT-6.1 Sol (55). Anthropic and OpenAI hold the top five.
- The best model with open weights, Kimi K3.1 Pro (weights not yet released), scores 46. Among models whose weights are already out, DeepSeek V4.1 Flash 1017 and GLM-5.3 lead with 32.
What it costs
Among the top ten, the cost of running one task from the tests differs by more than 13 times: from $0.96 (GPT-6.1 Terra) to $13.04 (Claude Fable 5.1). The ranking itself shows that the highest score is not always the best choice for a project.
How to read it
These are one company’s results, computed with its own method, and the rule change moved the standings, so I do not treat them as a final verdict on quality. They suggest instead that public tests are easy to train for. I write about choosing a model sensibly in the frontier models hub and about cheaper alternatives in the open models hub.