Kimi K3vsKimi K2.7 Code
Across 7 shared benchmarks, Kimi K3 leads overall: Kimi K3 wins 7, Kimi K2.7 Code wins 0, with 0 ties and an average score difference of +18.24.
Kimi K3
Moonshot AI · 2026-07-16 · Reasoning model
Kimi K2.7 Code
Moonshot AI · 2026-06-12 · Coding model
Kimi K37 wins(100%)(0%)0 winsKimi K2.7 Code
Benchmark scores
Grouped by capability, sorted by largest gap within each. 7 shared benchmarks.
Coding and Software Engineer
Kimi K3 4/4| Benchmark | Kimi K3 | Kimi K2.7 Code | Diff |
|---|---|---|---|
| DeepSWE | 67.505 / 26Max (With Tools) | 3123 / 26Normal (With Tools) | +36.50 |
| Program Bench | 77.801 / 5Max (With Tools) | 53.603 / 5Thinking (With Tools) | +24.20 |
| MLS Bench | 48.301 / 4Max (With Tools) | 35.103 / 4Thinking (With Tools) | +13.20 |
| Kimi Code Bench 2.0 | 72.901 / 3Max (With Tools) | 622 / 3Thinking (With Tools) | +10.90 |
AI Agent - Tool Usage
Kimi K3 3/3| Benchmark | Kimi K3 | Kimi K2.7 Code | Diff |
|---|---|---|---|
| Terminal-Bench 2.1 | 88.302 / 43Max (With Tools) | 67.0433 / 43Thinking (With Tools) | +21.26 |
| MCPMark-Verified | 94.501 / 3Max (With Tools) | 81.102 / 3Thinking (With Tools) | +13.40 |
| MCP-Atlas | 84.203 / 38Max (With Tools) | 7617 / 38Thinking (With Tools) | +8.20 |
Specs
| Field | Kimi K3 | Kimi K2.7 Code |
|---|---|---|
| Publisher | Moonshot AI | Moonshot AI |
| Release date | 2026-07-16 | 2026-06-12 |
| Model type | Reasoning model | Coding model |
| Architecture | MoE | MoE |
| Parameters | 2.8T | 1T |
| Context length | 1M | 256K |
| Max output | 1M | Not available |
API pricing
Prices use DataLearner records when available; missing fields are not inferred.
| Item | Kimi K3 | Kimi K2.7 Code |
|---|---|---|
| Text input | ¥20 / 1M tokens | $0.95 / 1M tokens |
| Text output | ¥100 / 1M tokens | $4 / 1M tokens |
| Cache read | ¥2 / 1M tokens | $0.19 / 1M tokens |
Summary
- Kimi K3leads in:Coding and Software Engineer (4/4), AI Agent - Tool Usage (3/3)
On average across the 7 shared benchmarks, Kimi K3 scores 18.24 higher.
Largest single-benchmark gap: DeepSWE — Kimi K3 67.50 vs Kimi K2.7 Code 31 (+36.50).
Page generated from structured model, pricing and benchmark records. No real-time LLM is used to write the prose.