OpenCode.json reasoningEffort setting doesnt work?

[mejobloggs-cw]

Description

I've tried a few different things similar to the config below, but every time I launch OpenCode it says Build mode is using GPT-5.4 HIGH

How can I set Build to medium via the config?

{
  "$schema": "https://opencode.ai/config.json",

  "provider": {
    "copilot": {}
  },

  "model": "github-copilot/gpt-5.4",
  "small_model": "github-copilot/gpt-5.4-mini",

  "agent": {
    "plan": {
      "model": "github-copilot/gpt-5.4",
      "reasoningEffort": "high",
      "textVerbosity": "low"
    },
    "build": {
      "model": "github-copilot/gpt-5.4",
      "reasoningEffort": "medium",
      "textVerbosity": "low"
    }
  }
}

Plugins

No response

OpenCode version

1.14.30

Steps to reproduce

Set config with

  "agent": {
    "plan": {
      "model": "github-copilot/gpt-5.4",
      "reasoningEffort": "high",
      "textVerbosity": "low"
    },
    "build": {
      "model": "github-copilot/gpt-5.4",
      "reasoningEffort": "medium",
      "textVerbosity": "low"
    }
  }

• Run opencode
• See if Build says "High" or "Medium"

Screenshot and/or share link

No response

Operating System

Latest Debian

Terminal

No response

[github-actions]

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

• OpenAI reasoningEffort=xhigh is not visible in TUI or export #18634: reasoningEffort set for the build agent is not visibly confirmed to be active in TUI or export — may reflect the same underlying problem where the setting is parsed but not applied
• subagent model variants are parsed but not applied at runtime in v1.4.0 #21632: Agent-level model config (including reasoningEffort) is parsed correctly but not applied at runtime in the subagent execution path

If your issue is specifically that the setting is being silently ignored (rather than just not displayed), #21632 in particular may describe the same root cause.

If you believe this is a distinct issue, feel free to clarify how your case differs from those reports.

[mejobloggs-cw]

If I'm doing the opencode.json wrong, is there a way to save current TUI settings?

I tried setting to Medium via the TUI which works, but I couldnt see any way to save/export the settings.

I'm not sure if those issues above are duplicates of my issue. I don't know enough of the opencode technicals to understand. I'm just a new user trying to setup a "starter config" with my team

[rekram1-node]

Try using variants instead (see docs)

[mejobloggs-cw]

How do you reference a variant? The docs dont seem to cover that

For example, this shows how to create "thinking" variant:
https://opencode.ai/docs/models/#custom-variants

Then i tried something like

"build": { "model": "github-copilot/gpt-5/thinking" }

But opencode says that's invalid.

I read through the whole models section and didn't spot a way.

[rekram1-node]

How are u trying to reference it? like whats your goal?

[mejobloggs-cw]

The end goal is to have Opencode use GPT 5.4 Medium by reading the setting from Opencode.json (not manually switching it to medium via the UI)

Following this example https://opencode.ai/docs/models/#custom-variants I attempted to create a "Thinking" variant (just following the example. If it worked I would then create a "Medium" variant)

That example creates the variant, but how do I set Build mode to use that variant? The docs don't appear to say how

I tried both of these (just attempting to mimic the json hierarchy hoping it would work), but they resulted in errors:

"build": { "model": "github-copilot/gpt-5/thinking" }

and

"build": { "model": "github-copilot/gpt-5", "variant": "thinking" }

My current Full opencode.json, using variants as suggested, which still shows in the UI as Build using GPT 5.4 High

{
    "$schema": "https://opencode.ai/config.json",
    "provider": {
        "copilot": {
            "models": {
                "gpt-5.4": {
                    "variants": {
                        "plan-high": {
                            "reasoningEffort": "high",
                            "textVerbosity": "low"
                        },
                        "build-medium": {
                            "reasoningEffort": "medium",
                            "textVerbosity": "low"
                        }
                    }
                }
            }
        }
    },
    "model": "github-copilot/gpt-5.4",
    "agent": {
        "plan": {
            "model": "github-copilot/gpt-5.4",
            "variant": "plan-high"
        },
        "build": {
            "model": "github-copilot/gpt-5.4",
            "variant": "build-medium"
        }
    }
}