Kimi K2.6
Moonshot AI · released 2026-04-01 · Modified MIT (commercial restrictions) license
Kimi K2.6 (Moonshot AI, April 2026) is a 1T-param open-weight MoE focused on agentic coding: 80.2% SWE-bench Verified, 89.6% LiveCodeBench, 90.5% GPQA, 96.4% AIME (vendor pipeline). License has commercial restrictions.
Key specs
| Type | Local open-weight |
|---|---|
| Parameters | — |
| Architecture | — |
| Context window | — |
| Knowledge cutoff | — |
| Modalities | text |
| Recommended backends | — |
| Minimum viable rig | — |
Benchmark scores
| GPQA Diamond | 90.5% |
|---|---|
| SWE-bench Verified | 80.2% |
| AIME | 96.4% |
| MMLU-Pro | — |
| BFCL v3 (tool use) | — |
| Composite score | 7.23 |
| Community rating | No reviews yet |
Strengths & weaknesses
Strengths: State-of-the-art open-weight coding/agentic model (SWE-bench Verified 80.2); Very strong reasoning: GPQA 90.5, AIME 96.4; 1T MoE (32B active), 262K context; Long-horizon agentic execution
Weaknesses: Modified-MIT license with commercial-use restrictions; Coding scores use Moonshot's in-house agent framework