release aidev SIG 2/5

opencode 1.17.9: GLM-5.2 high/max thinking variants, step-limit behavior fix

Small release. Agent step limits now end with a forced final text response instead of failing mid-run, GLM-5.2 gains high and max thinking variants across supported providers, and follow-up user messages are no longer wrapped in a steering reminder so prompt caching keeps working.

PUBLISHED2026-06-21
OBSERVED2026-08-11
AGE1mo
SOURCES1
  • Core: configured agent step limits force a final text response rather than failing mid-run
  • Core: high and max thinking variants added for GLM-5.2 across supported providers
  • Core: follow-up user messages no longer wrapped in a steering reminder, preserving prompt caching
  • Core: Devstral model detection fixed for provider IDs with differing casing
  • Core: configured custom headers now passed to Copilot model requests
  • TUI: background subagent shortcut only shown when the server supports it
fixmodel-supportfeature

COMMUNITY

No curated reactions recorded for this event. Facts and takes are kept in separate layers — community context is added by hand, never blended into the record above.