Artificial Analysis moved to Intelligence Index v5 on 27 Oct, using mostly private tests. Every score dropped: GLM-5.3 went from 45 to 32. The highest-scoring downloadable models score 32; the top closed model, Claude Opus 5.5, scores 58.
Scores quoted before 27 Oct do not compare with today's, so check which version a vendor is citing. For the hardest tasks, closed frontier models still score higher than any model you can download.
What happened
Artificial Analysis, an independent benchmarking firm, moved to version 5 of its Intelligence Index on 27 October 2026. The new version relies mostly on private tests.
Key facts
- Every score dropped. GLM-5.3 scored 45 under the previous version and 32 under v5.
- The top closed model, Claude Opus 5.5, scores 58.
- The highest-scoring released Open-weight model: A model whose weights the lab publishes, so you can download it and run it on your own hardware. Glossary models, DeepSeek V4.1-Flash-1017 and GLM-5.3, score 32: 26 points below the top closed model.
- Kimi K3.1 Pro scores 46, but its weights are not yet downloadable.
Our view
When a vendor quotes an Artificial Analysis score, ask which index version it is. A v4 number next to a v5 number is not a fair comparison. The new scores also show that, for the most demanding reasoning tasks, closed frontier models still score well above anything you can run yourself. For many routine business tasks, a smaller model you control may be enough; test on your own work before deciding.
Why it matters: Check which index version a vendor is quoting, because v4 and v5 scores do not compare.
Source: Artificial Analysis, “Announcing Artificial Analysis Intelligence Index v5”, 27 Oct 2026