Kimi K2.6
Kimi K2.6 is Moonshot AI's open-source, native multimodal agentic model focused on state-of-the-art coding, long-horizon execution, and agent swarm capabilities. It scales horizontally to 300 sub-agents executing 4,000 coordinated steps, dynamically decomposing tasks into parallel, domain-specialized subtasks. K2.6 unifies text, image, and video input with thinking and non-thinking modes, supports a 256K context, and powers proactive 24/7 background agents that manage schedules, execute code, and orchestrate cross-platform operations without human oversight.
Benchmark results
| Benchmark | Score | Tags | Source |
|---|---|---|---|
| AIME 2026 | 96.4% | self-reported llm-stats | link → |
| APEX-Agents | 27.9% | self-reported llm-stats | link → |
| BabyVision | 68.5% | self-reported llm-stats | link → |
| BrowseComp | 86.3% | self-reported llm-stats | link → |
| CharXiv-R | 86.7% | self-reported llm-stats | link → |
| Claw-Eval | 80.9% | self-reported llm-stats | link → |
| DeepSearchQA | 83.0% | self-reported llm-stats | link → |
| GPQA | 90.5% | self-reported llm-stats | link → |
| HMMT Feb 26 | 92.7% | self-reported llm-stats | link → |
| Humanity's Last Exam | 36.4% | self-reported llm-stats | link → |
| IMO-AnswerBench | 86.0% | self-reported llm-stats | link → |
| LiveCodeBench v6 | 89.6% | self-reported llm-stats | link → |
| MathVision | 93.2% | self-reported llm-stats | link → |
| MCP-Mark | 55.9% | self-reported llm-stats | link → |
| MMMU-Pro | 80.1% | self-reported llm-stats | link → |
| OJBench | 60.6% | self-reported llm-stats | link → |
| OSWorld-Verified | 73.1% | self-reported llm-stats | link → |
| SciCode | 52.2% | self-reported llm-stats | link → |
| SWE-bench Multilingual | 76.7% | self-reported llm-stats | link → |
| SWE-Bench Pro | 58.6% | self-reported llm-stats | link → |
| SWE-Bench Verified | 80.2% | self-reported llm-stats | link → |
| Terminal-Bench 2.0 | 66.7% | self-reported llm-stats | link → |
| Toolathlon | 50.0% | self-reported llm-stats | link → |
| V* | 96.9% | self-reported llm-stats | link → |
| WideSearch | 80.8% | self-reported llm-stats | link → |