Moonshot AI's Kimi K2.7 Code boosts efficiency with 30% fewer thinking tokens

Moonshot AI has made its latest coding model, Kimi K2.7 Code, available in public preview on Microsoft Foundry starting July 1, 2026. The model is specifically designed for complex, long-horizon coding workflows, multi-step execution, and agentic tasks — i.e., AI that can plan and execute sequences of actions autonomously. According to the company, Kimi K2.7 Code reduces 'thinking-token' usage by approximately 30% compared to its predecessor K2.6, while simultaneously achieving higher benchmark performance. This means developers can get better coding assistance with less wasted computation on intermediate reasoning steps.

The reduction in thinking tokens is particularly valuable for agentic coding scenarios, where the AI must reason through multiple steps (e.g., debugging, refactoring, testing) before producing final code. By cutting overhead, K2.7 Code can handle longer workflows within context limits and at lower cost. The public preview on Microsoft Foundry gives enterprise teams easy access via Azure infrastructure, including integration with existing CI/CD pipelines. Moonshot AI positions this as a practical tool for automating multi-file edits, dependency resolution, and complex logic synthesis. Early benchmarks suggest K2.7 Code outperforms K2.6 on standard coding benchmarks like HumanEval and SWE-bench, though exact numbers were not disclosed.