gpt-6-luna usage reported although never used #52367

Activity

1. github-actions

   added

   needs:complianceThis means the issue will auto-close after 2 hours.This means the issue will auto-close after 2 hours.

   and removed

   needs:complianceThis means the issue will auto-close after 2 hours.This means the issue will auto-close after 2 hours.

   on Sep 30, 2026
2. github-actions commented on Sep 30, 2026

   

   github-actionsBot

   on Sep 30, 2026 – with GitHub Actions

   Contributor

   More actions

   Thanks for updating the issue with the report details and system information!
3. github-actions commented on Sep 30, 2026

   

   github-actionsBot

   on Sep 30, 2026 – with GitHub Actions

   Contributor

   More actions

   Thanks for updating the issue with the report details and system information!
4. Dante-dan commented on Sep 30, 2026

   

   Dante-dan

   on Sep 30, 2026

   Contributor

   More actions

   OpenCode makes a separate background request to generate the session title. In the v2 code, the title selector picks a small model from the conversation's provider, and that selector prefers the gpt-luna family. The Go model catalog includes gpt-6-luna in that family. Selecting Space Bunny or Longcat for the conversation therefore does not prevent a separate GPT-6-Luna title request.

   The console records usage against the requested model, rather than replacing its name with the upstream response model. Background title generation is a plausible explanation for the small rows you describe. This is a code trace, not verification of your account's requests; matching those rows' session IDs and timestamps to new-session/title generation would confirm whether they are title traffic.
5. daddy33 commented on Oct 1, 2026

   

   daddy33

   on Oct 1, 2026

   Author

   More actions

   OpenCode makes a separate background request to generate the session title. In the v2 code, the title selector picks a small model from the conversation's provider, and that selector prefers the gpt-luna family. The Go model catalog includes gpt-6-luna in that family. Selecting Space Bunny or Longcat for the conversation therefore does not prevent a separate GPT-6-Luna title request.

   The console records usage against the requested model, rather than replacing its name with the upstream response model. Background title generation is a plausible explanation for the small rows you describe. This is a code trace, not verification of your account's requests; matching those rows' session IDs and timestamps to new-session/title generation would confirm whether they are title traffic.

   Thanks a lot for clarifying. But in my usage tab, I can read this for example:

   "requestedModel" "model" "provider" "cost"

   "gpt-6-luna" "gpt-6-luna" "opencode" "0.0001215"

   Obviously, the amount is ridiculously small, but does it mean that using a free model is not (completely) free?

   👍React with 👍1Reacted by serhumanos
6. serhumanos commented on Oct 2, 2026

   

   serhumanos

   on Oct 2, 2026

   More actions

   It would be great if we could manually pick the model used for title generation. For instance, I'd love to use MuseSpark1.3 since it has a more generous quota than GPT-Luna in my case. Being able to configure this would be helpful!
   thanks!
7. Dante-dan commented on Oct 2, 2026

   

   Dante-dan

   on Oct 2, 2026

   Contributor

   More actions

   That row records a nonzero cost for a GPT-6-Luna request. It doesn't identify what triggered that request, or establish that Space Bunny/Longcat itself incurred a charge. The default auxiliary-model selector prefers model families without checking their price; selecting a free conversation model therefore doesn't guarantee that automatically selected title traffic uses a free model. I can't confirm whether this particular row was title traffic without its session/time correlation. Can you check whether one of those rows matches a new session's timestamp/session ID?

   For manually choosing the title model, this setting already exists: set small_model in opencode.json to the exact provider/model-id shown by /models, using a model available to your account. See the configuration docs. In V2, it maps to the title agent's model, which takes precedence over automatic selection.
8. serhumanos commented on Oct 3, 2026

   

   serhumanos

   on Oct 3, 2026

   More actions

   For manually choosing the title model, this setting already exists: set small_model in opencode.json to the exact provider/model-id shown by /models, using a model available to your account. See the configuration docs. In V2, it maps to the title agent's model, which takes precedence over automatic selection.

   this works, I use the v2 setting:

     "agents": {
       "title": {
         "model": "opencode-go/muse-spark-1.3-contributor"
       },
     },

   now I only have muse spark 1.3 requests

   thank you very much!

   👍React with 👍1Reacted by ASID
9. neriousy commented on Oct 21, 2026

   

   neriousy

   on Oct 21, 2026

   Member

   More actions

   These are the background requests OpenCode makes to generate session titles, so your account hasn't been compromised. By design, titles use a small, cheap model from the provider you're using instead of your main model, and on OpenCode Go/Zen that's gpt-6-luna, which is why the amounts are so small.

   If you'd rather titles use a different model, set it in your config:

   { "agents": { "title": { "model": "opencode-go/<model-id>" } } }

   (small_model works too.) Closing as this is working as intended.
10. neriousy

    closed this as not plannedon Oct 21, 2026