Repository navigation
GLM-5 subagents freeze when using zai-coding-plan provider #14333
Description
Description
Having GLM-5 subagents from the zai-coding-plan provider randomly freeze before completing the task. This does not occur when running GLM-5 as the main agent -- so it's a specific incompatibility between the subagent tracking -- will need to diagnose what sort of issue it is.
Prompt caching issue possibly? -- it happens even when only using a single subagent, so that narrows it down a bit, maybe at longer contexts a subagent isn't properly handling prompt caching
Other issue -- how subagent provider requests are handled -- I know that there is some special handling for some providers for subagents, like for copilot. It appears that the same issue is not popping up in claude code, but that is using the anthropic endpoint.
This is preventing me from using glm-5 with subagent-driven systems -- I have run into similar issues with GLM-4.7 when using multiple subagents, but this could be a different issue
Different models/providers are working with subagents -- e.g. github-copilot/sonnet-4.6
This is NOT a problem that can be hand-waved away as a provider-related problem -- there is llkely a fix related to how we handle zai-coding-plan models when used as subagents.
Plugins
"opencode-qwencode-auth@latest", "openslimedit@latest"
OpenCode version
OpenCode 1.2.6
Steps to reproduce
- Use GLM-5 as main agent -- no issues, runs just fine
- Use Kimi-K2.5 as main agent, call GSD-related subagents running with providers other than zai-coding-plan -- no problem
- Use Kimi-K2.5 (or other model) as main agent, call subagents running with zai-coding-plan provider and GLM-4.7 or GLM-5 model -- subagent has a high chance of freezing (no more output from subagent, freezes the session, fails to complete after 30 minutes)
Screenshot and/or share link
No response
Operating System
Ubuntu 25.04 (served session), MacOS Tahoe 26.2 (attached opencode session)
Terminal
Kitty
Activity
github-actions commentedon Feb 19, 2026
This issue might be a duplicate of existing issues. Please check:
Your issue may be provider-specific (zai-coding-plan/GLM-5 interaction) rather than a general subagent timeout gap, but the underlying freeze-with-no-recovery behavior overlaps significantly with those reports. It's worth checking if any fixes from those issues would address this case, or if a separate provider-specific fix is needed.
github-actions commentedon Apr 21, 2026
To stay organized issues are automatically closed after 60 days of no activity. If the issue is still relevant please open a new one.