GitHub Copilot session auto-titling fails when enterprise policy disables gpt-5-mini #17531

Activity

1. johnnyhuy

   added

   bugSomething isn't workingSomething isn't working

   on Mar 14, 2026
2. github-actions

   assigned

   rekram1-node

   on Mar 14, 2026
3. github-actions

   added

   acp

   coreAnything pertaining to core functionality of the application (opencode server stuff)Anything pertaining to core functionality of the application (opencode server stuff)

   on Mar 14, 2026
4. github-actions commented on Mar 14, 2026

   

   github-actionsBot

   on Mar 14, 2026 – with GitHub Actions

   Contributor

   More actions

   This issue might be a duplicate of existing issues. Please check:

   • Show error message when session auto-naming fails due to disabled small model #9902: Same root cause and symptoms — session auto-naming silently fails when the small model is disabled by a provider/enterprise policy, with the same suggested fixes (fallback or surface an error).
   • [bug] Session auto-naming silently fails when small_model is invalid #14807: Same silent failure behavior when small_model is unavailable, with detailed source-level analysis and the same suggested resolution.

   If your issue adds new context specific to the GitHub Copilot enterprise policy scenario, please clarify how it differs from those issues.
5. johnnyhuy commented on Mar 14, 2026

   

   johnnyhuy

   on Mar 14, 2026

   Author

   More actions

   Happy to code commit something though there's a few considerations

   1. A warning message should suggest to configure small model(s) to fall back on
   2. If setting the global small model it sets it for all, which may inconsistently work based on provider availability
6. StefanLobbenmeier commented on Mar 30, 2026

   

   StefanLobbenmeier

   on Mar 30, 2026

   More actions

   Our enterprise allows usage of 4.1 and 4o but not 5-mini (not sure if there is a reason or if models are only enabled after someone requests them)

   Since all three of them are free to use, maybe we can automatically fallback to the other 2 in case it fails? Although to be honest I would still prefer a warning, but the warning could suggest to try one of the three and link to the copilot models to show which models are cheap: https://docs.github.com/en/copilot/reference/ai-models/supported-models#model-multipliers
7. t-monaghan commented on Mar 30, 2026

   

   t-monaghan

   on Mar 30, 2026

   More actions

   I've been having the same session title issues.
   Model provider - GitHub Copilot
   Small model - Haiku 4.5 (enabled)

   Interestingly 5-mini is enabled for my organisation, and after configuring the small model to 5-mini the issue has resolved.
8. SsparKluo

   marked title agent use not allowed param reasoning_effort: minimal #22796 as a duplicate of this issue on Apr 16, 2026
9. SsparKluo

   unmarked title agent use not allowed param reasoning_effort: minimal #22796 as a duplicate of this issue on Apr 16, 2026
10. mrw commented on Apr 28, 2026

    

    mrw

    on Apr 28, 2026

    More actions

    I ran into the same issue and asked OpenCode to investigate:

    What's happening

    1. OpenCode auto-selects github-copilot/gpt-5-mini as your small_model (per the priority list in the binary).
    2. For title generation, OpenCode sends reasoning_effort: "minimal" (it wants the cheapest/fastest call).
    3. GitHub Copilot's gpt-5-mini endpoint rejects "minimal" — it only accepts low, medium, or high. The request 400s.
    4. The error is swallowed silently (no retry, no fallback), so the session keeps its placeholder title New session - .
       This is a Copilot-side API change — gpt-5-mini used to accept minimal. Your DB confirms it: every session through April 12 got a real title; everything April 15 and later is stuck as "New session - ...".

    Fix
    Pin small_model to one that doesn't get a reasoning_effort: "minimal" rejection. Add to ~/.config/opencode/opencode.json:
    "small_model": "github-copilot/claude-haiku-4.5"

    I added "small_model": "github-copilot/claude-haiku-4.5", to ~/.config/opencode/opencode.json and it fixed it.
11. StefanLobbenmeier commented on Apr 29, 2026

    

    StefanLobbenmeier

    on Apr 29, 2026

    More actions

    Haiku has a multiplier of 0.33, I would recommend using one of the free models instead so you don’t burn your tokens on it:
    https://docs.github.com/en/copilot/reference/ai-models/supported-models#model-multipliers

    So GPT 4.1, 4o, 5 mini or Raptor mini
12. mrw commented on Apr 29, 2026

    

    mrw

    on Apr 29, 2026

    More actions

    @StefanLobbenmeier I hear you on that, but the issue here is that we can't use gpt-5-mini because opencode hardcodes reasoning_effort: "minimal" AFAIK, and copilot doesn't support that on gpt-5-mini anymore. I'm not sure about the others.
13. rekram1-node commented on Apr 29, 2026

    

    rekram1-node

    on Apr 29, 2026

    Collaborator

    More actions

    @mrw that is inaccurate, we dont hardcode that for copilot anymore
14. rekram1-node commented on Apr 29, 2026

    

    rekram1-node

    on Apr 29, 2026

    Collaborator

    More actions

    If you are on a more recent opencode version I think this should be completely closed. We sync the models from ur account now so if it's disabled it should use other models
15. rekram1-node

    closed this as completedon Apr 29, 2026