Moonshot AI open-sources Kimi K2.7 Code, cuts average token use 30%

The open-source coding model is available through APIs and on Hugging Face, a move that could broaden access to lower-cost AI coding tools for developers.

Summary

Moonshot AI released and open-sourced Kimi K2.7 Code, positioning the model as an upgrade for instruction following and long-context coding tasks. The company said the release reduces average token consumption by 30%, a change that could lower usage intensity for developers handling large prompts and extended code-generation workloads. It also reported code-performance gains of 21.8% on Kimi Code Bench v2 and about 10% on Kimi Claw 24/7 Bench. Kimi K2.7 Code is available through APIs and on Hugging Face, extending access to the model as Moonshot AI frames its efficiency and open-source design as a way to reduce costs and widen use of AI-driven coding tools.

Terms & Concepts
  • open-source: Software or models made publicly available for use, inspection and modification.
  • long-context coding: Using AI systems to work with very large prompts, files or codebases over extended inputs.
  • Hugging Face: A platform widely used to distribute and access open-source AI models and tools.