Across 17 shared benchmarks, Kimi K2.5 scores higher on 9 and muse-glimmer-30b on 8. The widest gap is HLE, where Kimi K2.5 scores 50.2 against 22. muse-glimmer-30b is the cheaper of the two on tracked API pricing ($0.30 against $0.56 per million input tokens).
| Benchmark | Kimi K2.5 | muse-glimmer-30b |
|---|---|---|
| AA-LCR | 73 | 80 |
| aime_2026 | 95.8 | 94.7 |
| baby_vision | 36.5 | 70.4 |
| charxiv_rq | 78.7 | 85.9 |
| deepsearchqa_f1 | 89 | 74.6 |
| GPQA Diamond | 87.9 | 83.5 |
| HLE | 50.2 | 22 |
| IFBench | 70.2 | 77 |
| MCP Atlas | 64 | 75.5 |
| MMMU-Pro | 78.5 | 74 |
| OmniDocBench 1.5 | 88.8 | 75.8 |
| OSWorld-Verified | 63.3 | 65.9 |
| scicode | 49 | 43.6 |
| SWE-bench Pro | 53.8 | 51.2 |
| SWE-bench Verified | 76.8 | 76 |
| Tau 3 Banking | 14.2 | 23.5 |
| Terminal-Bench 2.1 | 45.7 | 51.7 |
Best tracked score per model per benchmark (default configuration; source-attributed). ↓ marks lower-is-better metrics. Open either model for its full surface, provenance and pricing. Quoted rates are the price-setter row we currently track for each model — its direct or vendor-official listing where one exists (moonshot-official, deepinfra), otherwise the lowest tracked offer.