No reasoning at all through Github Copilot provider

Description

Validate thinking in VSCode

Attempt to get thinking in `opencode`

opencode.jsonc

{
  "$schema": "https://opencode.ai/config.json",
  "theme": "aura",
  "autoupdate": true,
  "provider": {
    "github-copilot": {
      "models": {
        "[Custom] Opus 4.5 Thinking": {
          "id": "claude-opus-4.5",
          "reasoning": true,
          "options": {
            "thinking": {
              "type": "enabled",
              "budgetTokens": 32000,
            },
          }
        },
      }
    }
  }
}

A conversation example:

opencode export

{
  "info": {
    "id": "ses_4753afc9affenaSxWVHstiDVAN",
    "version": "1.1.1",
    "projectID": "e53faf218247b17967230b742668df6addde9727",
    "directory": "/home/jcarlens/Development/sesm",
    "title": "Checking assistant reasoning",
    "time": {
      "created": 1767559725925,
      "updated": 1767559730475
    },
    "summary": {
      "additions": 0,
      "deletions": 0,
      "files": 0
    }
  },
  "messages": [
    {
      "info": {
        "id": "msg_b8ac50368001zOrnH6Xk5rnVK3",
        "sessionID": "ses_4753afc9affenaSxWVHstiDVAN",
        "role": "user",
        "time": {
          "created": 1767559725933
        },
        "summary": {
          "title": "Asking about reasoning",
          "diffs": []
        },
        "agent": "build",
        "model": {
          "providerID": "github-copilot",
          "modelID": "[Custom] Opus 4.5 Thinking"
        }
      },
      "parts": [
        {
          "id": "prt_b8ac50368002yxYgBf1OX3yO1R",
          "sessionID": "ses_4753afc9affenaSxWVHstiDVAN",
          "messageID": "msg_b8ac50368001zOrnH6Xk5rnVK3",
          "type": "text",
          "text": "Are you reasoning?"
        }
      ]
    },
    {
      "info": {
        "id": "msg_b8ac503730018iRxad0NY14TX9",
        "sessionID": "ses_4753afc9affenaSxWVHstiDVAN",
        "role": "assistant",
        "time": {
          "created": 1767559725939,
          "completed": 1767559730467
        },
        "parentID": "msg_b8ac50368001zOrnH6Xk5rnVK3",
        "modelID": "[Custom] Opus 4.5 Thinking",
        "providerID": "github-copilot",
        "mode": "build",
        "agent": "build",
        "path": {
          "cwd": "/home/jcarlens/Development/sesm",
          "root": "/home/jcarlens/Development/sesm"
        },
        "cost": 0,
        "tokens": {
          "input": 327,
          "output": 77,
          "reasoning": 0,
          "cache": {
            "read": 10534,
            "write": 0
          }
        },
        "finish": "stop"
      },
      "parts": [
        {
          "id": "prt_b8ac5101b001lc4JrFFJfRMDvj",
          "sessionID": "ses_4753afc9affenaSxWVHstiDVAN",
          "messageID": "msg_b8ac503730018iRxad0NY14TX9",
          "type": "step-start",
          "snapshot": "e0066eb511d26f679f2ba4739f14047f10f435aa"
        },
        {
          "id": "prt_b8ac5101c001ZUyiAP75JVK2pe",
          "sessionID": "ses_4753afc9affenaSxWVHstiDVAN",
          "messageID": "msg_b8ac503730018iRxad0NY14TX9",
          "type": "text",
          "text": "No, I'm not using an extended reasoning or \"thinking\" mode. I'm a standard Claude model responding directly to your messages. I process your input and generate responses, but I don't have a separate visible chain-of-thought or reasoning trace that I'm working through before answering.\n\nIs there something specific you'd like help with in your project?",
          "time": {
            "start": 1767559730442,
            "end": 1767559730442
          }
        },
        {
          "id": "prt_b8ac5150c001FbC86cMHHQDIoA",
          "sessionID": "ses_4753afc9affenaSxWVHstiDVAN",
          "messageID": "msg_b8ac503730018iRxad0NY14TX9",
          "type": "step-finish",
          "reason": "stop",
          "snapshot": "e0066eb511d26f679f2ba4739f14047f10f435aa",
          "cost": 0,
          "tokens": {
            "input": 327,
            "output": 77,
            "reasoning": 0,
            "cache": {
              "read": 10534,
              "write": 0
            }
          }
        }
      ]
    }
  ]
}

I also tried to put any value in the reasoning, nothing seems to break the API. Maybe the options are ignored?

Thanks for your hardwork!

Plugins

No response

OpenCode version

1.1.1

Steps to reproduce

Connect to github copilot provider
Either use built-in Opus 4.5 with any built-in variants, or try to configure a custom model
Ask anything to see a chain of thougth

Screenshot and/or share link

No response

Operating System

Ubuntu 24.04

Terminal

Ghostty

Note: I did the same kind of exercice with GPT 5.2 (with the reasoning effort settings), same issue.

I tried to run mitmproxy to see VSCode requests body.
Here is a snippet:

{
  "messages": [
    {
      "role": "system",
      "content": "You are an expert AI...",
      "copilot_cache_control": {
        "type": "ephemeral"
      }
    }
  ],
  "model": "claude-opus-4.5",
  "temperature": 0,
  "top_p": 1,
  "max_tokens": 16000,
  "tools": [],
  "n": 1,
  "stream": true,
  "thinking_budget": 15999
}

I removed some useless details.
It seems like it defines a thinking_budget field in the root of the JSON payload the the /completions endpoint. Interesting.

GPT 5.2 through copilot:

{
  "model": "gpt-5.2",
  "input": [
    {
      "role": "system",
      "content": []
    }
  ],
  "stream": true,
  "tools": [],
  "max_output_tokens": 64000,
  "store": false,
  "truncation": "disabled",
  "reasoning": {
    "effort": "high",
    "summary": "detailed"
  },
  "include": [
    "reasoning.encrypted_content"
  ]

We can see the reasoning and include at the root too.
From what I understood, this is because it uses the Responses API. While opencode use the completions one for my test prompts. OpenCode is apparently using Respones API only for codex according to my opencode thread.

I'm seeing the same issue. Would love to be able to configure thinking/reasoning for Copilot models.

This is a critical bug if wanting to use copilot provider as there is a huge difference for models like gpt 5.2 with reasoning on/off.

Also would like to know if this is getting implemented.

same for me

Would be very helpful to have it since Anthropic subscription is not an option anymore.

so is it actually reasoning but we dont see the thinking block or its not reasoning at all?

@hazrid93

@hazrid93 no reasoning at all.

Hope this gets fixed

@thdxr

@thdxr hope this get priority

Responses reasoning fixed, next we need to do completions

Has this issue been resolved?

I'm getting the same issue, but with GCP Vertex as a provider for Opus 4.5. Is anyone else experiencing that?

@MichaelYochpaz

@MichaelYochpaz that would be separate and specific to gcp, same thing happens even w/ ctrl+t?

@rekram1-node

@rekram1-node Thanks for the response. Yep, I tried switching modes, it didn't make a difference.

I asked here because that's the only related issue I could find, and the fact that no one else opened an issue about this made me doubt that maybe it's just me doing something wrong (I'm a new OpenCode user).

But regardless, I opened a new issue specifically for Vertex - #9894

Tested and reasoning is working with GPT-5.2 but unsure how to configure reasoning/thinking for Opus 4.5 through Github copilot.

@jbuckmccready

@jbuckmccready that's because completions hasn't been fixed yet, see here

@MichaelYochpaz

@MichaelYochpaz mind opening a separate issue then?

@jbuckmccready that's because completions hasn't been fixed yet, see here

I've been using copilot-api for quite some time now. this branch maybe helpful for fixing this issue.

#10092 seems interesting

looks like gemini 3 pro and opus still don't show thinking?

anthropic models should all work in next release, now just need to merge in that linked pr to get the google models working, but openai and anthropic shoudl both work fine w/ reasoning (in next release)

Unfortunately the messages api wasn't ready for us, going to fix it on completions api instead for now, people were getting 429s due to this change^ had to revert

Merging the pr tonight, also I learned that gemini models are expected not to have variants, so really it's just the anthropic models (cause openai ones work already)

Navigation Menu

Repository navigation

No reasoning at all through Github Copilot provider #6864

Description

Validate thinking in VSCode

Attempt to get thinking in `opencode`

Plugins

OpenCode version

Steps to reproduce

Screenshot and/or share link

Operating System

Terminal

8 remaining items

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

No reasoning at all through Github Copilot provider #6864

Description

Description

Validate thinking in VSCode

Attempt to get thinking in opencode

Plugins

OpenCode version

Steps to reproduce

Screenshot and/or share link

Operating System

Terminal

Activity

Coruscant11 commented on Jan 4, 2026

Coruscant11 commented on Jan 4, 2026

Coruscant11 commented on Jan 4, 2026

NicolaiOksen commented on Jan 7, 2026

jbuckmccready commented on Jan 9, 2026

pushad commented on Jan 12, 2026

dreambit commented on Jan 12, 2026

iamim commented on Jan 13, 2026

hazrid93 commented on Jan 16, 2026

Coruscant11 commented on Jan 16, 2026

doriuf commented on Jan 16, 2026

hazrid93 commented on Jan 16, 2026

8 remaining items

rekram1-node commented on Jan 19, 2026

hjnnjh commented on Jan 21, 2026

MichaelYochpaz commented on Jan 21, 2026

rekram1-node commented on Jan 21, 2026

MichaelYochpaz commented on Jan 21, 2026

jbuckmccready commented on Jan 21, 2026

rekram1-node commented on Jan 21, 2026

rekram1-node commented on Jan 21, 2026

hjnnjh commented on Jan 22, 2026

Coruscant11 commented on Jan 22, 2026

amat27 commented on Jan 23, 2026

rekram1-node commented on Jan 26, 2026

rekram1-node commented on Jan 29, 2026

rekram1-node commented on Jan 31, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

Attempt to get thinking in `opencode`