Infinite compaction loop on every response even with empty session #31152

Activity

1. github-actions

   assigned

   rekram1-node

   on Jun 6, 2026
2. github-actions commented on Jun 6, 2026

   

   github-actionsBot

   on Jun 6, 2026 – with GitHub Actions

   Contributor

   More actions

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

   • Infinite Build ↔ Compaction loop after task completion #31111: Infinite Build ↔ Compaction loop after task completion (very similar symptom)
   • OpenCode immediately enters auto-compaction loop and stops generating responses #30680: OpenCode immediately enters auto-compaction loop and stops generating responses (same behavior on fresh session)
   • bug(session): infinite compaction loop when compression fails to reduce context #27924: bug(session): infinite compaction loop when compression fails to reduce context (same loop pattern)
   • OpenCode immediately triggers “Input exceeds context window” and auto-compaction in empty sessions after latest update #27315: OpenCode immediately triggers compaction in empty sessions after latest update (similar empty-session scenario)

   If your issue is distinct from all of the above, please add a comment explaining how it differs so maintainers can keep it open.
3. jiten-core commented on Jun 7, 2026

   

   jiten-core

   on Jun 7, 2026

   More actions

   this is exaly same thing happend with me
4. bfanger commented on Jun 18, 2026

   

   bfanger

   on Jun 18, 2026

   More actions

   I had this issue too when using a local modal, in my case I was able to solve it by specifing the input limit.

   "models": {
     "Qwen3.6-27B-Fast": {
       "name": "Qwen3.6 27B Fast (MTP 40K)",
       "limit": {
         "context": 40000,
         "input": 40000,
         "output": 32000
       }
     },

   Could be related to #13980
5. tavurth commented on Jun 26, 2026

   

   tavurth

   on Jun 26, 2026

   Last edited by tavurth

   More actions

   Interesting that this is happening to me too, but seems to be model dependant.

   MiMo V2.5 (openrouter/xiaomi/mimo-v2.5) started doing this, but swapping to another model still works.

   Uninstalling and reinstalling openrouter does not fix.

   Neither does removing the entire ~/.local/share/openrouter, ~/.config/openrouter and ~/.cache/opencode folders.

   (MiMo was previously working just fine)

   Fixed:

   ~/.config/opencode/config.json

   {
     "provider": {
       "openrouter": {
         "models": {
           "xiaomi/mimo-v2.5": {
             "limit": {
               "context": 1000000,
               "output": 1000000
             }
           }
         }
       }
     }
   }

   openrouter was returning incorrect context for that model. Upstream bug.
6. SMohsenM commented on Jul 1, 2026

   

   SMohsenM

   on Jul 1, 2026

   More actions

   @tavurth I don’t think it’s an upstream bug. Here are the results from https://openrouter.ai/api/v1/model/xiaomi/mimo-v2.5 . The DigitalOcean provider has a 32K context limit and is currently the top provider.

   {
       "data": {
           "id": "xiaomi/mimo-v2.5",
           "canonical_slug": "xiaomi/mimo-v2.5-20260422",
           "hugging_face_id": "XiaomiMiMo/MiMo-V2.5",
           "name": "Xiaomi: MiMo-V2.5",
           "created": 1776874269,
           "description": "MiMo-V2.5 is a native omnimodal model by Xiaomi. It delivers Pro-level agentic performance at roughly half the inference cost, while surpassing MiMo-V2-Omni in multimodal perception across image and video understanding...",
           "context_length": 1048576,
           "architecture": {
               "modality": "text+image+audio+video->text",
               "input_modalities": [
                   "text",
                   "audio",
                   "image",
                   "video"
               ],
               "output_modalities": [
                   "text"
               ],
               "tokenizer": "Other",
               "instruct_type": null
           },
           "pricing": {
               "prompt": "0.000000105",
               "completion": "0.00000028"
           },
           "top_provider": {
               "context_length": 32000,
               "max_completion_tokens": null,
               "is_moderated": false
           },
           "per_request_limits": null,
           "supported_parameters": [
               "frequency_penalty",
               "include_reasoning",
               "logit_bias",
               "logprobs",
               "max_tokens",
               "presence_penalty",
               "reasoning",
               "repetition_penalty",
               "response_format",
               "seed",
               "stop",
               "structured_outputs",
               "temperature",
               "tool_choice",
               "tools",
               "top_k",
               "top_logprobs",
               "top_p"
           ],
           "default_parameters": {
               "temperature": 1,
               "top_p": 0.95,
               "top_k": null,
               "frequency_penalty": null,
               "presence_penalty": null,
               "repetition_penalty": null
           },
           "supported_voices": null,
           "knowledge_cutoff": null,
           "expiration_date": null,
           "links": {
               "details": "/api/v1/models/xiaomi/mimo-v2.5-20260422/endpoints"
           },
           "benchmarks": {
               "design_arena": [
                   {
                       "arena": "models",
                       "category": "3d",
                       "elo": 1301,
                       "win_rate": 54.3,
                       "rank": 18
                   },
                   {
                       "arena": "models",
                       "category": "asciiart",
                       "elo": 1194,
                       "win_rate": 48.3,
                       "rank": 25
                   },
                   {
                       "arena": "models",
                       "category": "codecategories",
                       "elo": 1304,
                       "win_rate": 55,
                       "rank": 14
                   },
                   {
                       "arena": "models",
                       "category": "dataviz",
                       "elo": 1280,
                       "win_rate": 53.8,
                       "rank": 16
                   },
                   {
                       "arena": "models",
                       "category": "gamedev",
                       "elo": 1301,
                       "win_rate": 55.8,
                       "rank": 16
                   },
                   {
                       "arena": "models",
                       "category": "svg",
                       "elo": 1219,
                       "win_rate": 52.3,
                       "rank": 21
                   },
                   {
                       "arena": "models",
                       "category": "uicomponent",
                       "elo": 1312,
                       "win_rate": 56.2,
                       "rank": 12
                   },
                   {
                       "arena": "models",
                       "category": "website",
                       "elo": 1304,
                       "win_rate": 55.2,
                       "rank": 12
                   }
               ]
           },
           "reasoning": {
               "mandatory": false
           }
       }
   }
   
7. tavurth commented on Jul 1, 2026

   

   tavurth

   on Jul 1, 2026

   More actions

   @SMohsenM the MiMo V2.5 context limit is actually 1M tokens, and so it's an upstream issue.

   Specifically, it's an issue with the Output tokens limit vs the Context tokens limit.

   A second diagnostic utility here is that it stopped working without upgrading Opencode. So one day it worked, the next day not. I.e. upstream provider issue.

   I dug fairly deeply through the codebase, and there is a way to actively bypass this, but the config change recommended in #31152 (comment) is the better short term approach.

   However, I would suggest opencode show an error dialog in such cases as the passed JSON model config has some undefined values in this case, which could be caught, and then a human input box could be allowed.
8. github-actions commented on Aug 31, 2026

   

   github-actionsBot

   on Aug 31, 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.
9. github-actions

   closed this as not plannedon Aug 31, 2026