Kimi K2.5 (Thinking): benchmark results
Kimi K2.5 evaluated with thinking enabled. Provider: Moonshot. Released 2026-01-27. Access: Open.
Aggregate Rating 1697 ± 5, rank #367 of 2168 rated models, from 537 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| GISA - Item EM | 68.18 | Exact match (%) on the 22 item-type queries (a single answer | 100 |
| Synthetic Hospital - Patient Diagnosis | 73.2 | Severity-weighted F1 (%; reconstruction of the patient longi | 100 |
| TIDE (Evolving Documents, Gold Context) - Context Misalignment | 27.62 | Strict accuracy (%; 116 questions whose preamble states a su | 100 |
| TIDE (Evolving Documents, Gold Context) - Event Sorting | 66.73 | Strict accuracy (%; chronological ordering of 207 shuffled s | 100 |
| WildTableBench - Aggregate Verification | 81.8 | Accuracy (%) on the Aggregate Verification subtype (C3-A), 9 | 100 |
| BenchTable - Utility | 87.4 | Weighted Score (%) | 99.6 |
| AA Omniscience - Software Engineering (SWE) - Rust | 78 | Accuracy (%) | 97.6 |
| WildTableBench - Semantic Lookup and Structure | 56.7 | Accuracy (%) on the Semantic Lookup and Structure subtype (C | 97.5 |
| AgentProcessBench - tau2-Bench First-Error Accuracy | 65.6 | FirstErrAcc (%): share of trajectories where the judge locat | 97.4 |
| AA Omniscience - Software Engineering (SWE) - TypeScript | 60 | Accuracy (%) | 97.3 |
| AA Omniscience - Software Engineering (SWE) - HTML | 66 | Accuracy (%) | 96.8 |
| AA Omniscience - Software Engineering (SWE) - Python | 47 | Accuracy (%) | 96.6 |
Interactive version: theaggregate.ai/model?slug=kimi-k2-5-thinking · How It Works · Data refreshed daily, snapshot 2026-10-20.