0T params
Total model parameters
0B active
Active per token (MoE)
0K tokens
Context window
0
Expert modules
| K2.5 Score | Context | |
|---|---|---|
| SWE-Bench Verified | 76.8 | Outperforms Gemini Pro 3.1 |
| SWE-Bench Multilingual | 73.0 | Outperforms GPT-5.2 and Gemini Pro 3.1 |
| AIME 2025 | 96.1 | Advanced math competition problems |
| GPQA Diamond | 87.6 | Graduate level science questions |
| HLE (with tools) | 50.2 | At 76% lower cost than Claude Opus 4.6 |
| LiveCodeBench | 85.0 | Real time coding evaluation |
| MathVision | 84.2 | Mathematical visual reasoning |
| OCRBench | 92.3 | Optical character recognition accuracy |
| VideoMMMU | 86.6 | Video understanding and reasoning |
| WeirdML | 46% | Unusual/edge case reasoning (lower than competitors) |
| Speed | Best For | |
|---|---|---|
| Instant | Fastest | Quick questions, simple tasks, casual conversation |
| Thinking | Medium | Math, logic, complex analysis, coding problems |
| Agent | Slower | Research, multi-step tasks, file processing |
| Agent Swarm (beta) | Variable | Deep research, large codebase analysis, comprehensive investigations |
Why MoE matters for cost
With only 32 billion parameters active per token out of 1 trillion total, K2.5 achieves frontier performance while using a fraction of the compute per inference. This is why API pricing can be $0.60/$3.00 per million tokens while matching models that cost 5x to 25x more.
K2.5 in third party tools: the Cursor controversy
In early 2026, it was revealed that Cursor uses Kimi K2.5 in its Composer 2 feature. This sparked debate about model attribution and transparency, with users questioning whether AI coding tools should clearly disclose which underlying models power their features. On the positive side, it validates K2.5's coding capabilities at the frontier level, since a leading AI code editor chose it for a core feature.