Moonshot AI has officially released and open-sourced Kimi-K2.7-Code, its latest model focused on coding and agentic workflows. The new release reportedly delivers significant gains over K2.6 across multiple coding benchmarks while also reducing reasoning-token usage by around 30%. Some of the reported improvements include: • +21.8% on Kimi Code Bench v2 • +11.0% on Program Bench • +31.5% on MLS Bench Lite • 30% lower reasoning-token usage And honestly, that’s what makes this release interesting. Raw benchmark scores matter, but efficiency matters too. A model that solves more problems while using fewer tokens can be faster, cheaper, and more practical for real-world coding tasks. Moonshot AI also says K2.7-Code improves instruction following and long-horizon reasoning, helping it handle complex multi-step software engineering workflows with higher success rates. The bigger story is how competitive the open-source coding model space is becoming. Just a few years ago, the strong...