← comparisons / kimi-k2.7-code-vs-kimi-k3
Kimi K2.7 Code vs Kimi K3
01Head to head
| Kimi K2.7 Code | Kimi K3 | |
|---|---|---|
| Creator | Moonshot AI | Moonshot AI |
| Released | 2026-06-12 | 2026-07-16 |
| Context | 262K | 1M |
| Max output | 262K | 131K |
| Input $/M | — | — |
| Output $/M | — | — |
| Knowledge cutoff | 2025-01 | — |
| Modalities | text + image + video → text | text + image + video → text |
| Capabilities | tool-use, extended-thinking, structured-output | tool-use, extended-thinking, structured-output |
| Providers | Moonshot AI (Kimi) | Moonshot AI (Kimi) |
02Benchmarks · vendor-reported unless verified
| Kimi K2.7 Code | Kimi K3 | |
|---|---|---|
| DeepSWE | — | 67.5% |
| FrontierSWE (dominance score) | — | 81.2 |
| GDPval-AA (Elo) | — | 1668 |
| Kimi Claw 24/7 Bench | 46.9% | — |
| Kimi Code Bench | 62% | — |
| MCP Atlas (success rate) | 76 | — |
| MCP Mark Verified (success rate) | 81.1 | — |
| MLS Bench Lite | 35.1% | — |
| Program Bench | 53.6% | — |
| Program Bench (score) | — | 77.8 |
| SWE Marathon | — | 42% |
| Terminal-Bench (accuracy) | — | 88.3 |