Tag
Stepwise AI is giving away tokens for free, but users speculate its purpose is to collect code for training its own model. Current quality is mediocre and the terms are unfriendly.
This paper introduces LoopCoder-v2, a 7B code model that benefits most from a single rethinking loop; additional loops degrade performance, challenging the assumption that more test-time compute always helps.
Moonshot AI releases Kimi K2.7 Code, a 1T parameter Mixture-of-Experts model focused on coding and agentic tasks, with improved token efficiency and strong benchmark results against GPT-5.5 and Claude Opus 4.8.