Currently experiencing halted and slow inference in OpenCode Go #34667

Activity

1. github-actions

   assigned

   StarpTech

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

   

   github-actionsBot

   on Jun 30, 2026 – with GitHub Actions

   Contributor

   More actions

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

   • Main & Sub-agents Randomly Freeze Indefinitely: Frontend Permanently Shows "thinking" with No Errors, While Actual LLM Inference Has Already Terminated Prematurely #24342: Similar report of DeepSeek V4 halting/freezing mid-generation on Windows, with the frontend stuck showing "thinking" while inference has already terminated
   • enters into constant thinking loop.... #29512: Similar report of OpenCode entering a constant thinking/halt state after an update

   If your issue is the same, consider adding a thumbs-up or comment there instead. If it's distinct (e.g. specific to the models you mentioned or a recent regression), feel free to keep this open and clarify the difference.
3. raymelon commented on Jul 5, 2026

   

   raymelon

   on Jul 5, 2026

   Author

   More actions

   Same issue again. Just now. Same model. MiMo V2.5 with internal server errors, and once it pushes through, cache is busted at 0.
4. GusBot69 commented on Jul 15, 2026

   

   GusBot69

   on Jul 15, 2026

   More actions

   Same issue — consistent slowness through Hermes Agent

   I'm experiencing the same slowness with MiMo V2.5 through OpenCode Go, though my symptoms are slightly different from the halting issue — I get consistent 2-4.5 second latency per request rather than complete halts.

   Setup:

   • Client: Hermes Agent (not OpenCode CLI)
   • Model: mimo-v2.5 via OpenCode Go
   • Region: US

   Speed test results (July 15, 2026):

   • Run 1: 2.2s, 35 tokens (14 reasoning)
   • Run 2: 2.3s, 31 tokens (15 reasoning)
   • Run 3: 4.5s, 50 tokens (49 reasoning)

   The hidden reasoning tokens (~15-49 per request) add overhead that cannot be disabled through the Go endpoint, even with thinking: {"type": "disabled"} in extra_body.

   For comparison, MiMo V2.5 direct from Xiaomi's API runs at 85 t/s. Through OpenCode Go it's effectively 10-20 t/s after proxy overhead.

   This is not just a CLI issue — it affects any OpenAI-compatible client hitting the Go endpoint.
5. github-actions commented on Sep 14, 2026

   

   github-actionsBot

   on Sep 14, 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 Sep 14, 2026