Kimi K2.7 Code
by Moonshot AI
Coding-specialized variant of K2.6 (June 2026). +21.8% improvement on coding benchmarks vs K2.6. 30% fewer reasoning tokens per agentic loop. Production-ready for software engineering tasks.
Release
- Date: June 2026
- Parameters: 1T total / 32B active
- Focus: Coding and software engineering
Specialization
- Coding improvement: +21.8% on internal benchmarks vs K2.6
- Token efficiency: 30% fewer reasoning tokens per agentic loop
- Architecture: K2.6 foundation + coding-optimized fine-tuning
Performance
- Advanced code generation and debugging
- Long-horizon coding tasks
- Complex multi-file refactoring
- No independent leaderboard validation at launch
Pricing
- Input: $0.60 per million tokens
- Output: $2.50 per million tokens
- Same as K2.6 (specialist variant pricing)
Use Cases
- Production software engineering
- Autonomous code review and refactoring
- Complex programming problems
- Multi-language development
Positioning
- Alternative to K2.6 for coding-heavy workloads
- Cost-effective vs K3 for specialized coding
- Higher efficiency on domain-specific tasks
See Also
2026 update
| Date | Event |
|---|---|
| 2026-06-12 | Released — tested as possibly the best open coding model |
| 2026-06-17 | Reported to beat Opus 4.8 and GPT-5.5, “really cheap” |
| 2026-06-18 → 06-25 | Broad practitioner adoption; Obsidian and NotebookLM pairings |
| 2026-07-07 | First open-weight model available in GitHub Copilot |
What changed
K2.7 Code is the high-water mark of the Kimi line’s visibility. The GitHub Copilot inclusion
(2026-07-07) is the substantive milestone — the first open-weight model offered inside Microsoft’s
mainstream coding product, which matters beyond benchmark claims.
Scope note: this page covers the Code variant specifically. For the wider generation see
Kimi K2 series.