Devin adds Kimi K2.7 Code, GLM 5.2

// 1h agoPRODUCT UPDATE

Devin adds Kimi K2.7 Code, GLM 5.2

Devin now supports Kimi K2.7 Code and GLM 5.2 in its desktop and CLI environments, bringing two top-tier open-weight models to Western developer workflows. Both models perform competitively against GPT-5.5 on the FrontierCode Extended engineering benchmark.

// ANALYSIS

The integration of leading Chinese models into a flagship US coding agent signals a shift in the frontier landscape.

–GLM 5.2 (43.0%) and Kimi K2.7 Code (39.5%) are now viable alternatives to GPT-5.5 and Claude Opus for real-world engineering
–GLM's massive 1M context window makes it uniquely suited for large-scale repository refactoring
–Kimi's native visual reasoning provides strong capabilities for UI and frontend implementation tasks
–Offering free tier access lowers the barrier for developers to test non-US models in production environments

// TAGS

devinkimiglmai-codingcoding-agentllmopen-weightsagent

DISCOVERED

1h ago

2026-06-25

PUBLISHED

8h ago

2026-06-24

RELEVANCE

9/ 10

AUTHOR

devindesktop

// KEEP READING

More AI developer news from the feed

EXPLORE FULL FEED

UPDATE3h ago

Cursor runs coding agents from CI

Cursor introduces remote, VM-backed background agents that can be triggered directly from CI pipelines and persist through local network disconnections. The agents run asynchronously in isolated cloud sandboxes, allowing developers to offload long-running tasks and receive completed pull requests hours later.

NEWS4h ago

Tesana user builds playable Backrooms game

A creator leveraged Tesana's prompt-to-world AI engine to build a playable Backrooms game following the release of the new Backrooms movie. The project demonstrates the platform's ability to rapidly generate topical 3D experiences without traditional game development.

NEWS6h ago

LuaJIT 3.0 proposes modern syntax extensions

Mike Pall has proposed a set of modern syntax extensions for LuaJIT 3.0, introducing features like nil-coalescing, optional chaining, and compound assignment. These features aim to improve developer quality-of-life and will be backported to LuaJIT 2.1 to ease compiler bootstrapping.