OpenHands 1.34 adds per-conversation Docker runtimes, runtime settings, automation tag editing and custom-script model selection. It also moves a session API key from process arguments to the environment, preserves automation KV data across key rotation and blocks unsafe markdown targets.
Why it matters: The release improves workload isolation and automation recoverability while closing concrete secret-transport and link-opening gaps—controls that matter more as coding agents keep long-lived state and execute unattended work.
DataClawEval provides 100 Docker-isolated tasks across PySpark, MySQL, HiveSQL, Trino and FlinkSQL. Agents inspect schemas, implement pipelines, execute them and materialize artifacts that deterministic scripts grade; the strongest reported configuration scored 74.9.
Why it matters: Text-to-SQL scores omit most of the work required to build and maintain real data systems, including execution, artifacts and engine-specific behavior.
Moonshot AI released Kimi-Dev-72B for software issue resolution, trained with reinforcement learning in Docker repositories and rewarded when the full test suite passed. Weights and code were published through Hugging Face and GitHub.
Why it matters: The training objective moved closer to a verifiable engineering outcome than next-token prediction or model-judged code style.