OpenAI reasoningEffort=xhigh is not visible in TUI or export

[Valeron2206]

Summary

When reasoningEffort is set explicitly for an OpenAI build agent, OpenCode does not visibly expose whether the setting is actually active.

Config used

{
  "$schema": "https://opencode.ai/config.json",
  "agent": {
    "build": {
      "model": "openai/gpt-5.4",
      "reasoningEffort": "xhigh",
      "textVerbosity": "low",
      "reasoningSummary": "auto"
    }
  }
}

What I expect

At least one reliable way to confirm the active reasoning setting for the session, for example:

• visible in the TUI model/status line
• present in opencode export
• visible through a config/debug endpoint

What happens instead

• TUI only shows Build · gpt-5.4
• opencode export <session> shows providerID: openai and modelID: gpt-5.4
• there is no obvious variant, reasoningEffort, or equivalent field to confirm that xhigh is active

Why this matters

OpenCode docs document OpenAI reasoning variants (low, medium, high, xhigh) and also show direct agent-level reasoning options, but in practice there is no trustworthy runtime signal that the configured reasoning effort was actually applied.

This makes it hard to distinguish between:

1. the setting is applied but not displayed
2. the setting is ignored

Reproduction

1. Set agent.build.model = openai/gpt-5.4
2. Set agent.build.reasoningEffort = xhigh
3. Start a new session
4. Observe TUI header / export output

Actual session behavior observed

• new session clearly uses openai/gpt-5.4
• no visible indication of xhigh

Request

Please either:

• surface active reasoning settings in TUI/export/debug output, or
• document the canonical way to verify that reasoningEffort is being applied at runtime

[github-actions]

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

• Web UI does not reflect configured reasoning effort from model options #17588: Web UI does not reflect configured reasoning effort from model options — same root problem (configured reasoningEffort not surfaced in UI/output), though that report targets the web UI while this one targets the TUI and opencode export

If you believe these are distinct enough to warrant separate tracking, please let a maintainer know.

[Valeron2206]

Yes, this looks like the same underlying problem as #17588.

My report covers a slightly different surface area:

• TUI and opencode export, not only web UI
• agent-level config (agent.build.reasoningEffort = xhigh), not only model-level provider options

But the root issue seems the same: the effective reasoning setting is not surfaced clearly enough in runtime/UI/export, so it is impossible to verify whether it was applied.

If maintainers prefer to track this under #17588, I am fine with closing this as a duplicate.