Models · Compare
Kimi K2.5 vs Claude Opus 4.6
Rows highlighted in warm gray are where the models differ. Numbers carry their as-of date and primary source.
Specs
| Field | A: Kimi K2.5 | B: Claude Opus 4.6 |
|---|---|---|
| Released | 2026-01-27 | 2026-02-05 |
| Developer | Moonshot AI | Anthropic |
| Openness | Open | Proprietary |
| License | Modified MIT | Proprietary |
| OSI-approved | no | no |
| Data released | no | no |
| Training code | no | no |
| Architecture | moe | unknown |
| Total params | 1T | — |
| Active params | 32B | — |
| Experts | 384 (8 active) | — |
| Context window | 256K | 1.0M |
| Attention | mla | unknown |
| Position enc. | rope | unknown |
| Pretraining tokens | 15.0T | — |
| Post-training | sft, rlhf | rlhf, constitutional |
| Training hardware | — | — |
| $/M input | $0.58 | $5.00 |
| $/M output | $3.00 | $25.00 |
| Output tok/sec | 33.9 | 43.7 |
Benchmarks
Missing scores render as not reported; never inferred. Bold highlights the leader per benchmark.
General reasoning
| GPQA-Diamond | 87.6 2026-01-27 | 84.0 2026-05-21 |
Code
| SWE-Bench Verified | 76.8 2026-01-27 | — |
| LiveCodeBench | 85.0 2026-01-27 | — |
Math
| AIME 2025 | 96.1 2026-01-27 | — |
Context · A
Native multimodal Kimi at 1T total / 32B active with MoonViT (400M-parameter vision encoder), trained on roughly 15T mixed visual and text tokens. 256K context, Modified MIT license, January 27 2026 release. Adds visual reasoning, video understanding, and UI-to-code workflows.
Context · B
First Opus-class model with a 1M-token context window (in beta), released February 5 2026. Headline feature was agent teams in Claude Code, letting multiple agents work in parallel and coordinate autonomously. Long-context pricing premium kicks in above 200K input tokens ($10 / $37.50 per Mtok).