Persistent invalid reasoning value ("low") overrides env and CLI settings

[xuxiaoyang-if]

Description

Bug Description

When running opencode with gpt-5.2-chat-latest, the CLI fails with:
Error: Unsupported value: ‘low’ is not supported with the ‘gpt-5.2-chat-latest’ model.
Supported values are: ‘medium’.

This occurs even when:

• OPENCODE_REASONING=medium is exported
• No agent config sets low
• Binary is reinstalled
• Issue reproduces in both CLI and VS Code

---

Root Cause (Identified)

The invalid reasoning value (low) was persisted in local state directories:
~/.local/share/opencode
~/.config/opencode

Removing these directories resolves the issue immediately:

rm -rf ~/.local/share/opencode
rm -rf ~/.config/opencode

Expected Behavior
	•	Invalid or unsupported reasoning values should not silently persist
	•	Cached state should not override environment variables or CLI defaults
	•	A warning or validation should occur when loading incompatible state

Environment
	•	OS: macOS (Apple Silicon)
	•	Shell: zsh
	•	opencode version: 1.1.30
	•	Install method: curl installer

### Plugins

_No response_

### OpenCode version

1.1.30

### Steps to reproduce

_No response_

### Screenshot and/or share link

_No response_

### Operating System

macOS 26.2

### Terminal

kitty

[github-actions]

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

• Session Todo-list is forgotten after compaction #5934: Session todo-list is forgotten after compaction (related state persistence issue)
• Config variables in opencode.json overwritten with actual values on start up #9086: Config variables overwritten with actual values on startup (config state override issue)

The root cause appears to be related to how cached state persists and overrides environment variables or CLI defaults. Other related issues that discuss configuration persistence and validation:

• [FEATURE]: Persist desktop/web steps panel collapse state across sessions #7174: Persist desktop/web steps panel collapse state across sessions
• [FEATURE]: slash command for reload #6719: Slash command for reload

Feel free to ignore if your specific issue about reasoning values is distinct from these configuration override problems.

[Luketto8]

This issue persists in my case even after deleting said files.
Only problem appears to be chatgpt as some free models seem to function properly

[trogers1]

Fwiw, I tried deleting everything opencode-related that I could find that seemed like it could potentially have something cached for opencode:

rm -rf ~/.cache/opencode \
rm -rf ~/.local/share/opencode \
rm -rf ~/.local/share/opentui/opencode \
rm -rf ~/.local/state/opencode \
rm -rf ~/.config/opencode \
rm -rf ~/.opencode \
rm -rf ~/.bun/install/cache/@opencode-ai \
brew uninstall opencode

Then re-install, re-authenticate with OpenAI (via API Token)--Issue still persists.

[R44VC0RP]

Can you try running opencode upgrade @trogers1 as well as opencode export > session.json so I can see the session output?

[trogers1]

@R44VC0RP I had already uninstall+reinstalled so the upgrade command didn't do anything. But here's the exported session.json (with very minor edits for security):

{
  "info": {
    "id": "ses_3ef6cd3eaffeMDaSwUTXAWz6Hk",
    "slug": "misty-cabin",
    "version": "1.1.40",
    "projectID": "global",
    "directory": "/Code/debug",
    "title": "System plan mode: read-only analysis and delegation",
    "time": {
      "created": 1769804606485,
      "updated": 1769804606626
    },
    "summary": {
      "additions": 0,
      "deletions": 0,
      "files": 0
    }
  },
  "messages": [
    {
      "info": {
        "id": "msg_c10932c18001j49NhjsK5q9nTp",
        "sessionID": "ses_3ef6cd3eaffeMDaSwUTXAWz6Hk",
        "role": "user",
        "time": {
          "created": 1769804606493
        },
        "summary": {
          "title": "Text-only input: brief note \"Test\"",
          "diffs": []
        },
        "agent": "plan",
        "model": {
          "providerID": "openai",
          "modelID": "gpt-5.2-chat-latest"
        }
      },
      "parts": [
        {
          "id": "prt_c10932c180023XvXLt3PzJSCRp",
          "sessionID": "ses_3ef6cd3eaffeMDaSwUTXAWz6Hk",
          "messageID": "msg_c10932c18001j49NhjsK5q9nTp",
          "type": "text",
          "text": "Test"
        }
      ]
    },
    {
      "info": {
        "id": "msg_c10932c2d002s9sPM8svmgEbyg",
        "sessionID": "ses_3ef6cd3eaffeMDaSwUTXAWz6Hk",
        "role": "assistant",
        "time": {
          "created": 1769804606509,
          "completed": 1769804606783
        },
        "error": {
          "name": "APIError",
          "data": {
            "message": "Unsupported value: 'low' is not supported with the 'gpt-5.2-chat-latest' model. Supported values are: 'medium'.",
            "statusCode": 400,
            "isRetryable": false,
            "responseHeaders": {
              "alt-svc": "h3=\":443\"; ma=86400",
              "cf-cache-status": "DYNAMIC",
              "cf-ray": "9c63afe78f997851-SJC",
              "connection": "keep-alive",
              "content-length": "250",
              "content-type": "application/json",
              "date": "Fri, 30 Jan 2026 20:23:26 GMT",
              "openai-organization": "HIDDEN",
              "openai-processing-ms": "56",
              "openai-project": "HIDDEN",
              "openai-version": "2020-10-01",
              "server": "cloudflare",
              "strict-transport-security": "max-age=31536000; includeSubDomains; preload",
              "x-content-type-options": "nosniff",
              "x-request-id": "req_2c5d841c1a2c4699bdd82f83f0e37c95",
              "set-cookie": "HIDDEN"
            },
            "responseBody": "{\n  \"error\": {\n    \"message\": \"Unsupported value: 'low' is not supported with the 'gpt-5.2-chat-latest' model. Supported values are: 'medium'.\",\n    \"type\": \"invalid_request_error\",\n    \"param\": \"text.verbosity\",\n    \"code\": \"unsupported_value\"\n  }\n}",
            "metadata": {
              "url": "https://api.openai.com/v1/responses"
            }
          }
        },
        "parentID": "msg_c10932c18001j49NhjsK5q9nTp",
        "modelID": "gpt-5.2-chat-latest",
        "providerID": "openai",
        "mode": "plan",
        "agent": "plan",
        "path": {
          "cwd": "/Code/debug",
          "root": "/"
        },
        "cost": 0,
        "tokens": {
          "input": 0,
          "output": 0,
          "reasoning": 0,
          "cache": {
            "read": 0,
            "write": 0
          }
        }
      },
      "parts": []
    }
  ]
}

[R44VC0RP]

Confirmed this, was able to repro. 👍🏻

[Luketto8]

Did some light digging and the following caught my eye. Might be worth checking out.

gpt-5.2 modelID worked for me,
but gpt-5.2-chat-latest does not.

So appears to misbehave at

      if (
        input.model.api.id.includes("gpt-5.") &&
        !input.model.api.id.includes("codex") &&
        input.model.providerID !== "azure"
      ) {
        result["textVerbosity"] = "low"
      }