kimi-k3.1-pro
Meant to excel at
-
Scores 55 on the Artificial Analysis Intelligence Index (v4.3.2), fourth of 221 models on its model page and 11 points above Kimi K3 on the same version.
Independent evaluator artificialanalysis.ai -
A post-trained version of Kimi K3 with substantially larger post-training compute; Moonshot reports claim-level hallucination falling from 23.35% for K3 to 6.85%.
Vendor's own site platform.kimi.ai
Watch out for
-
Slow and not cheap per task on Artificial Analysis: 46.3 output tokens per second (182nd of 192) and $2.11 per Intelligence Index task (92nd of 121).
Independent evaluator artificialanalysis.ai -
Thinking is always on (effort low, high or max, with max the default), so it cannot be run as a non-reasoning model.
Vendor's own site platform.kimi.ai -
Weights are not public yet: Moonshot says the full open weights arrive by Nov 4.
Vendor's own site platform.kimi.ai
Benchmarks
Overall percentile 0.78
| Benchmark | Score | Percentile | Source | Date |
|---|---|---|---|---|
| artificial analysis intelligence index | 55 | 0.78 | Independent evaluator artificialanalysis.ai | 2026-10-26 |
Percentile is the share of tracked models this model beats on that benchmark, averaged across the group. Only benchmarks at least four tracked models share are scored.
Practitioner reports
No reports yet.
