Activity

1. github-actions

   assigned

   kitlangton

   on May 5, 2026
2. github-actions commented on May 5, 2026

   

   github-actionsBot

   on May 5, 2026 – with GitHub Actions

   Contributor

   More actions

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

   • OpenCode Desktop v1.2.25 broke previously working Codex models after upgrade (gpt-5.3-codex-spark unsupported, gpt-5.3-codex hangs) #17309: gpt-5.3-codex hangs / stops responding after OpenCode upgrade (same symptom: Codex model not responding)
   • Agent gets stuck and stops responding in Windows 11 IDE #16593: Agent gets stuck and stops responding in Windows 11 (same OS, same symptom)
   • Message sending is blocked in WSL on Windows (OpenCode version 1.4.4+). #22580: Message sending is blocked in WSL on Windows (same environment: WSL on Windows, messages not processing)

   If your issue is different from the above, please add more detail about what you observe (any error messages, logs, or screenshots) to help distinguish it.
3. VitalyM commented on Jun 13, 2026

   

   VitalyM

   on Jun 13, 2026

   More actions

   Workaround: empty/no response with Azure gpt-5.3-codex via the Responses API

   If opencode (or any client) hangs/returns nothing with gpt-5.3-codex on Azure while simple curl calls work, it's an Azure-side issue with one specific parameter combination:

   reasoning.effort = "medium" + function tools in the request → the model fails with:

   {"error":{"code":"unknown","message":"The model was unable to complete inference due to an internal error."}}

   (HTTP 503 for non-streaming; a response.failed event for streaming, which clients often swallow silently → you see no output.)

   I verified it deterministically by replaying the exact request body and only varying the effort:

   reasoning effort	with tools
   low	✅ works
   medium	❌ fails (503 / "unable to complete inference")
   high	✅ works

   Removing the tools, or removing the reasoning block, also makes medium work — so the trigger is specifically medium effort + tools. The default for the build/coding agent is medium, which is why it fails every time while title generation (no tools) succeeds.

   Fix

   Force the reasoning effort to low or high for this model. In opencode config:

   "models": {
     "gpt-5.3-codex": {
       "name": "gpt-5.3-codex",
       "options": { "reasoningEffort": "high" }   // "low" also works; just not "medium"
     }
   }

   Note: "minimal" returns HTTP 400 on this deployment — use low or high.

   This looks like an Azure deployment-side regression (a client downgrade made no difference), so it may resolve itself server-side, but the config override is a reliable workaround in the meantime.
4. TimChinye commented on Jun 24, 2026

   

   TimChinye

   on Jun 24, 2026

   More actions

   This is likely unrelated to #32606 and #32149 as the issue occurs regardless of when model is being used.
5. github-actions commented on Aug 24, 2026

   

   github-actionsBot

   on Aug 24, 2026 – with GitHub Actions

   Contributor

   More actions

   To stay organized issues are automatically closed after 60 days of no activity. If the issue is still relevant please open a new one.
6. github-actions

   closed this as not plannedon Aug 24, 2026