[BUG]: GitHub Copilot Responses API routing for GPT-5+

[christso]

Problem

When using the github-copilot provider, GPT-5+ models are currently routed through the Chat API. This prevents users from enabling Responses-API-only controls (eg. reasoningEffort, reasoningSummary) for those models, even when they are supported by the model and useful for agent behavior tuning.

This is especially noticeable for github-copilot/gpt-5.* where users expect to be able to configure reasoning parameters similarly to other OpenAI-compatible providers.

Proposed Solution

Add an opt-in provider option to route supported Copilot models through the Responses API:

• New config flag: provider.github-copilot.options.useResponsesApi: true
• Behavior:
  • Always route codex* models through sdk.responses(modelID) (existing behavior)
  • If useResponsesApi is enabled and the model is GPT-5 or later, route through sdk.responses(modelID)
  • Otherwise, default to sdk.chat(modelID) (no behavior change by default)

Example config:

{
  "provider": {
    "github-copilot": {
      "options": {
        "useResponsesApi": true
      }
    }
  }
}

Future (Deprecation)

Once GitHub Copilot / VS Code enables Responses API usage broadly (the VS Code setting is currently labeled Experimental), this option can likely be deprecated and/or defaulted on for supported models to reduce configuration surface area.

References

Current status of supported API route for provider github-copilot:

Model	API Route
gpt-4.1	chat
gpt-4o	chat
gpt-5	responses
gpt-5-codex	responses
gpt-5-mini	chat
gpt-5.1	responses
gpt-5.1-codex	responses
gpt-5.1-codex-max	responses
gpt-5.1-codex-mini	responses
gpt-5.2	responses

• VS Code UI (settings) relevant to this feature:
  • GitHub > Copilot > Chat: Use Responses Api (Experimental)
  • GitHub > Copilot > Chat: Responses Api Reasoning Effort (Experimental)
  • GitHub > Copilot > Chat: Responses Api Reasoning Summary (Experimental)

Screenshot of the VS Code settings panel showing the above Copilot settings.

Environment

• Repo: sst/opencode (dev)
• Platform observed: Windows (but the behavior is provider/model routing specific, not OS-specific)

[github-actions]

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

• github copilot: gpt-5-mini not supported via responses api #4943: github copilot: gpt-5-mini not supported via responses api

Feel free to ignore if none of these address your specific case.

[jbuckmccready]

Is this something Github changed recently? I had reasoningEffort working with GPT 5.2 and now it no longer works (no change to configuration on my end).

  "provider": {
    "github-copilot": {
      "models": {
        "gpt-5.2": {
          "options": {
            "reasoningEffort": "medium",
            "textVerbosity": "low",
            "reasoningSummary": "auto",
            "include": ["reasoning.encrypted_content"]
          }
        }
      }
    }

[christso]

Is this something Github changed recently? I had reasoningEffort working with GPT 5.2 and now it no longer works (no change to configuration on my end).

@jbuckmccready This is a change in OpenCode, not GitHub.

Commit f1bb587 (merged December 1st) broke reasoning support for GPT-5.2 and other non-Codex GPT-5 models. The change incorrectly assumed only Codex models support reasoning, so it only routes models with "codex" in the name to the API that supports reasoningEffort. Everything else goes to a different API that doesn't support it.

GPT-5.2 was working with reasoning before this commit. There's a PR to fix it.

@rekram1-node should this be reclassified from feature to a bug? I previously thought it was a feature until I discovered it was working previously.

[SamPeng87]

这是 Github 最近更改的东西吗？我在使用 GPT 5.2 时 reasoningEffort 一直可用，但现在不再可用了（我的配置没有任何更改）。

@jbuckmccready这是 OpenCode 的更改，不是 GitHub 的。

Commit（于 12 月 1 日合并）破坏了对 GPT-5.2 及其他非 Codex GPT-5 模型的推理支持。该更改错误地假定只有 Codex 模型支持推理，因此仅将名称中包含 "codex" 的模型路由到支持 reasoningEffort 的 API。其它所有模型都被路由到不支持该功能的不同 API。

GPT-5.2 在此提交之前的推理功能是正常的。这里有一个 用来修复它。

@rekram1-node这应该从 feature 重新归类为 bug 吗？我之前以为这是一个 feature，直到我发现它之前是正常工作的。

I think your PR fix still requires the above parameters to activate the thinking mode. Is my understanding correct?

[christso]

I think your PR fix still requires the above parameters to activate the thinking mode. Is my understanding correct?

@SamPeng87 Yes, your understanding is correct. This PR fixes the error that occurs when reasoning is enabled, but you still need to configure the parameters (reasoningEffort, reasoningSummary, include) in your opencode.json to activate thinking mode.