Model Details
Kimi K2.7 Code
Quick answer: Kimi K2.7 Code is Moonshot AI's most capable coding model, designed for reliable instruction following in long-context scenarios and high-success-rate completion of complex programming tasks. It is part of the Kimi K2 model family and sits above K2.6 in coding-specific capability. On LHTB it scored 0.367 mean reward (rank 9), solving 3 of 46 long-horizon terminal tasks.
At a Glance
Where Kimi K2.7 Code leads
- Coding-specialized performance — highest coding success rate in the Kimi K2 model family.
- Reliable instruction following in long-context programming scenarios.
- Competitive LHTB result of 0.367 — ahead of several higher-cost models.
Where it lags
- Primarily available via the Kimi platform API (Chinese pricing in CNY); global availability may be limited.
- General-purpose capability is narrower than the full K2.6 base model.
Best for: Agentic coding pipelines, long-horizon software engineering tasks, and instruction-intensive code generation.
What Kimi K2.7 Code Is
Kimi K2.7 Code is Moonshot AI's latest coding-specialised variant in the Kimi K2 model family, released in July 2026. While the K2.6 base model focuses on general intelligence and autonomous agent execution, K2.7 Code is purpose-built for software engineering: it prioritises accurate long-context instruction following, stable multi-step code generation, and high task-completion rates on complex programming challenges.
The model continues Moonshot AI's track record in coding. The K2 family has been integrated into tools such as Tencent Cloud's CodeBuddy and various developer platforms. K2.7 Code represents the latest step in that lineage, with improvements specifically targeting the gap between single-file code generation and sustained, multi-step agentic coding workflows.
On the Long-Horizon Terminal-Bench (LHTB), which measures agent performance on hard 90-minute terminal tasks, Kimi K2.7 Code scored 0.367 mean reward — placing 9th among 21 models and solving 3 of 46 tasks outright. This is consistent with the model's positioning as a strong coding specialist rather than a general-purpose frontier model.
Specifications
| Field | Value |
|---|---|
| Modalities | Text |
| Type | Coding specialist |
| Family | Kimi K2 |
| License | Proprietary (API only) |
| API model name | kimi-k2.7-code |
| Pricing (CNY) | Input ¥6.50 / MTok · Output ¥27.00 / MTok · Cache hit ¥1.30 / MTok |
Related Benchmarks
- Long-Horizon Terminal-Bench (LHTB) — 0.367 mean reward, rank 9 (3/46 solved)
- SWE-bench Verified — industry standard for coding agents
Benchmark Leaderboards
This model isn’t on any benchmark leaderboard yet.