OpenCode is unable to invoke third-party models added by enterprises in GitHub Copilot.

[KTY810]

Description

1. I have a GitHub Copilot Enterprise account, and our company has added third-party models that are not part of Copilot to the platform.
2. When I connect Copilot to OpenCode, I am unable to read and use the third-party models that have been added to Copilot.

Plugins

oh-my-openagent

OpenCode version

1.17.10

Steps to reproduce

1. Copilot Enterprise Administrator Adding Third-Party Models
2. Connecting OpenCode to Copilot
3. Viewing Copilot Models Recognized by OpenCode

Screenshot and/or share link

No response

Operating System

ubuntu22.04

Terminal

No response

[github-actions]

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

• [FEATURE]: dynamic model list for copilot plugin #13470: Feature request for dynamic model list for copilot plugin, which discusses the same root problem — enterprise admins can enable/disable models but OpenCode uses a static list from models.dev and doesn't query per-user available models.
• Copilot Business/Enterprise users can't use OpenCode — three compounding auth/endpoint/header issues #20759: Covers broader Copilot Business/Enterprise auth and endpoint issues, including model access problems for enterprise users.

If your issue is specifically about the static model list not reflecting enterprise-managed models, #13470 is likely the most relevant existing report.

[KTY810]

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

• [FEATURE]: dynamic model list for copilot plugin #13470: Feature request for dynamic model list for copilot plugin, which discusses the same root problem — enterprise admins can enable/disable models but OpenCode uses a static list from models.dev and doesn't query per-user available models.
• Copilot Business/Enterprise users can't use OpenCode — three compounding auth/endpoint/header issues #20759: Covers broader Copilot Business/Enterprise auth and endpoint issues, including model access problems for enterprise users.

If your issue is specifically about the static model list not reflecting enterprise-managed models, #13470 is likely the most relevant existing report.

There is a difference. What I'm focusing on is using third-party models configured in GitHub Copilot's enterprise settings. The current issue is that OpenCode cannot read them.

[KTY810]

OpenCode GitHub Copilot Experimental Models Debug Report

Issue Summary

OpenCode 1.17.11 does not automatically discover some GitHub Copilot Business / enterprise Experimental models that are visible to the same authenticated account when the request uses VS Code / Copilot CLI style headers.

The missing models are returned by the Copilot /models API only when these headers are included:

Editor-Version: vscode/1.107.1
Editor-Plugin-Version: copilot-chat/0.26.7
Copilot-Integration-Id: copilot-developer-cli

With OpenCode's default model discovery headers, the models are absent. Manually registering the models in opencode.json and adding the same headers at the model level makes OpenCode able to use them.

Environment

• OpenCode version: 1.17.11
• OS: Linux
• GitHub Copilot auth in OpenCode: oauth configured for provider github-copilot
• Copilot plan/SKU: Copilot Business seat, login redacted
• OpenCode config path: ~/.config/opencode/opencode.json

Reproduction Data

The same OpenCode GitHub Copilot OAuth token was used for both requests below. Only headers changed. Authorization values are redacted and not shown.

Request A: OpenCode-style default headers

Request target:

GET https://api.business.githubcopilot.com/models

Headers used, excluding authorization:

{
  "User-Agent": "opencode/1.17.11",
  "X-GitHub-Api-Version": "2026-06-01",
  "Accept": "application/json"
}

Sanitized response summary:

{
  "status": 200,
  "totalReturned": 33,
  "pickerEnabled": 16,
  "experimental": [],
  "allPickerIds": [
    "claude-haiku-4.5",
    "claude-opus-4.5",
    "claude-opus-4.6",
    "claude-opus-4.7",
    "claude-opus-4.8",
    "claude-sonnet-4.5",
    "claude-sonnet-4.6",
    "gemini-2.5-pro",
    "gemini-3-flash-preview",
    "gemini-3.1-pro-preview",
    "gemini-3.5-flash",
    "gpt-5-mini",
    "gpt-5.3-codex",
    "gpt-5.4",
    "gpt-5.4-mini",
    "gpt-5.5"
  ]
}

Request B: VS Code / Copilot CLI-style headers

Request target:

GET https://api.business.githubcopilot.com/models

Headers used, excluding authorization:

{
  "User-Agent": "GithubCopilot/1.155.0",
  "Editor-Version": "vscode/1.107.1",
  "Editor-Plugin-Version": "copilot-chat/0.26.7",
  "Copilot-Integration-Id": "copilot-developer-cli",
  "X-GitHub-Api-Version": "2026-06-01",
  "Accept": "application/json"
}

Sanitized response summary:

{
  "status": 200,
  "totalReturned": 39,
  "pickerEnabled": 21,
  "experimental": [
    "*/GLM/glm-5.1",
    "*/GLM/glm-5.2",
    "*/HUAWEI/glm-5.2",
    "*/QDeepseekV4/deepseek-v4-flash",
    "*/QDeepseekV4/deepseek-v4-pro",
    "*/Qwen37plus/qwen3.7-plus"
  ]
}

Experimental Model API Metadata

The Experimental models returned by Request B are picker-enabled and support /chat/completions.

[
  {
    "id": "*/QDeepseekV4/deepseek-v4-flash",
    "name": "deepseek-v4-flash",
    "model_picker_enabled": true,
    "supported_endpoints": ["/chat/completions"],
    "limits": {
      "max_context_window_tokens": 1000000,
      "max_prompt_tokens": 840000,
      "max_output_tokens": 128000
    },
    "supports": {
      "streaming": true,
      "tool_calls": true
    }
  },
  {
    "id": "*/QDeepseekV4/deepseek-v4-pro",
    "name": "deepseek-v4-pro",
    "model_picker_enabled": true,
    "supported_endpoints": ["/chat/completions"],
    "limits": {
      "max_context_window_tokens": 1000000,
      "max_prompt_tokens": 840000,
      "max_output_tokens": 128000
    },
    "supports": {
      "tool_calls": true
    }
  },
  {
    "id": "*/Qwen37plus/qwen3.7-plus",
    "name": "qwen3.7-plus",
    "model_picker_enabled": true,
    "supported_endpoints": ["/chat/completions"],
    "limits": {
      "max_context_window_tokens": 1000000,
      "max_prompt_tokens": 991000,
      "max_output_tokens": 64000
    },
    "supports": {
      "tool_calls": true
    }
  },
  {
    "id": "*/GLM/glm-5.1",
    "name": "glm-5.1",
    "model_picker_enabled": true,
    "supported_endpoints": ["/chat/completions"],
    "limits": {
      "max_context_window_tokens": 200000,
      "max_prompt_tokens": 167232,
      "max_output_tokens": 32768
    },
    "supports": {
      "tool_calls": true
    }
  },
  {
    "id": "*/GLM/glm-5.2",
    "name": "glm-5.2",
    "model_picker_enabled": true,
    "supported_endpoints": ["/chat/completions"],
    "limits": {
      "max_context_window_tokens": 1000000,
      "max_prompt_tokens": 868928,
      "max_output_tokens": 131072
    },
    "supports": {
      "tool_calls": true
    }
  },
  {
    "id": "*/HUAWEI/glm-5.2",
    "name": "huawei-glm5.2",
    "model_picker_enabled": true,
    "supported_endpoints": ["/chat/completions"],
    "limits": {
      "max_context_window_tokens": 192000,
      "max_prompt_tokens": 128000,
      "max_output_tokens": 64000
    },
    "supports": {
      "tool_calls": true
    }
  }
]

OpenCode Behavior Before Workaround

opencode models github-copilot

Before manual configuration, OpenCode listed only the standard GitHub Copilot models and did not list any */... Experimental model.

Direct use of a raw Experimental ID without registration

Command:

opencode run -m "github-copilot/*/GLM/glm-5.2" --format json "Reply exactly OK"

Observed result:

{
  "type": "error",
  "error": {
    "name": "UnknownError",
    "data": {
      "message": "Unexpected server error. Check server logs for details."
    }
  }
}

Relevant sanitized log:

timestamp=2026-06-26T07:57:28.626Z level=INFO message=created id=<session> version=1.17.11 directory=<home> model=undefined

This indicates that an unknown slash-containing model ID is not cleanly resolved when it is not already registered in OpenCode's model catalog.

Manual raw ID registration without model-level headers

After registering the raw model ID in a temporary local opencode.json, OpenCode did send the full model ID:

timestamp=2026-06-26T08:02:58.781Z level=INFO message=stream providerID=github-copilot modelID=*/GLM/glm-5.2 session.id=<session> small=false agent="Sisyphus - ultraworker" mode=primary
timestamp=2026-06-26T08:02:58.794Z level=INFO message="llm runtime selected" llm.runtime=ai-sdk llm.provider=github-copilot llm.model=*/GLM/glm-5.2

But the API returned:

{
  "error": {
    "message": "The requested model is not supported.",
    "code": "model_not_supported",
    "param": "model",
    "type": "invalid_request_error"
  },
  "metadata": {
    "url": "https://api.githubcopilot.com/chat/completions"
  }
}

This was fixed by adding the VS Code / Copilot CLI-style headers at the model level.

Confirmed Workaround

Register the Experimental models manually under the github-copilot provider and attach the VS Code / Copilot CLI-style headers to each model.

After this workaround, OpenCode lists the models:

github-copilot/*/GLM/glm-5.1
github-copilot/*/GLM/glm-5.2
github-copilot/*/HUAWEI/glm-5.2
github-copilot/*/QDeepseekV4/deepseek-v4-flash
github-copilot/*/QDeepseekV4/deepseek-v4-pro
github-copilot/*/Qwen37plus/qwen3.7-plus

And this command succeeds:

opencode run -m "github-copilot/*/GLM/glm-5.2" --format json "Reply exactly OK"

Sanitized successful response:

{
  "type": "text",
  "text": "OK"
}

Relevant sanitized success log:

timestamp=2026-06-26T08:14:20.350Z level=INFO message=stream providerID=github-copilot modelID=*/GLM/glm-5.2 session.id=<session> small=false agent="Sisyphus - ultraworker" mode=primary
timestamp=2026-06-26T08:14:20.358Z level=INFO message="llm runtime selected" llm.runtime=ai-sdk llm.provider=github-copilot llm.model=*/GLM/glm-5.2

Current Local OpenCode Configuration Applied

File changed:

~/.config/opencode/opencode.json

The current configuration adds these models under provider.github-copilot.models.

Each model has:

{
  "family": "custom",
  "tool_call": true,
  "temperature": false,
  "experimental": true,
  "headers": {
    "Editor-Version": "vscode/1.107.1",
    "Editor-Plugin-Version": "copilot-chat/0.26.7",
    "Copilot-Integration-Id": "copilot-developer-cli"
  }
}

The six configured model IDs are:

*/QDeepseekV4/deepseek-v4-flash
*/QDeepseekV4/deepseek-v4-pro
*/Qwen37plus/qwen3.7-plus
*/GLM/glm-5.1
*/GLM/glm-5.2
*/HUAWEI/glm-5.2

Copy-Paste Configuration Snippet

Use this shape in ~/.config/opencode/opencode.json under the top-level provider object. Merge it with existing providers instead of replacing the entire file.

{
  "provider": {
    "github-copilot": {
      "models": {
        "*/QDeepseekV4/deepseek-v4-flash": {
          "name": "deepseek-v4-flash",
          "family": "custom",
          "limit": { "context": 1000000, "input": 840000, "output": 128000 },
          "tool_call": true,
          "temperature": false,
          "experimental": true,
          "headers": {
            "Editor-Version": "vscode/1.107.1",
            "Editor-Plugin-Version": "copilot-chat/0.26.7",
            "Copilot-Integration-Id": "copilot-developer-cli"
          }
        },
        "*/QDeepseekV4/deepseek-v4-pro": {
          "name": "deepseek-v4-pro",
          "family": "custom",
          "limit": { "context": 1000000, "input": 840000, "output": 128000 },
          "tool_call": true,
          "temperature": false,
          "experimental": true,
          "headers": {
            "Editor-Version": "vscode/1.107.1",
            "Editor-Plugin-Version": "copilot-chat/0.26.7",
            "Copilot-Integration-Id": "copilot-developer-cli"
          }
        },
        "*/Qwen37plus/qwen3.7-plus": {
          "name": "qwen3.7-plus",
          "family": "custom",
          "limit": { "context": 1000000, "input": 991000, "output": 64000 },
          "tool_call": true,
          "temperature": false,
          "experimental": true,
          "headers": {
            "Editor-Version": "vscode/1.107.1",
            "Editor-Plugin-Version": "copilot-chat/0.26.7",
            "Copilot-Integration-Id": "copilot-developer-cli"
          }
        },
        "*/GLM/glm-5.1": {
          "name": "glm-5.1",
          "family": "custom",
          "limit": { "context": 200000, "input": 167232, "output": 32768 },
          "tool_call": true,
          "temperature": false,
          "experimental": true,
          "headers": {
            "Editor-Version": "vscode/1.107.1",
            "Editor-Plugin-Version": "copilot-chat/0.26.7",
            "Copilot-Integration-Id": "copilot-developer-cli"
          }
        },
        "*/GLM/glm-5.2": {
          "name": "glm-5.2",
          "family": "custom",
          "limit": { "context": 1000000, "input": 868928, "output": 131072 },
          "tool_call": true,
          "temperature": false,
          "experimental": true,
          "headers": {
            "Editor-Version": "vscode/1.107.1",
            "Editor-Plugin-Version": "copilot-chat/0.26.7",
            "Copilot-Integration-Id": "copilot-developer-cli"
          }
        },
        "*/HUAWEI/glm-5.2": {
          "name": "huawei-glm5.2",
          "family": "custom",
          "limit": { "context": 192000, "input": 128000, "output": 64000 },
          "tool_call": true,
          "temperature": false,
          "experimental": true,
          "headers": {
            "Editor-Version": "vscode/1.107.1",
            "Editor-Plugin-Version": "copilot-chat/0.26.7",
            "Copilot-Integration-Id": "copilot-developer-cli"
          }
        }
      }
    }
  }
}


## Commands Used for Verification

List GitHub Copilot models visible to OpenCode:

```bash
opencode models github-copilot

Run one Experimental model:

opencode run -m "github-copilot/*/GLM/glm-5.2" --format json "Reply exactly OK"

Expected minimal response:

OK

Suggested Upstream Fix Areas

1. GitHub Copilot model discovery should use the same headers official Copilot clients use, or otherwise preserve account-specific enterprise / Experimental model visibility.
2. provider.github-copilot.options.headers appears in resolved config, but in this test it did not make opencode models github-copilot discover the Experimental models. If provider-level headers are intended to affect Copilot model discovery, this may be a bug.
3. If a slash-containing model ID is not registered, opencode run -m provider/model/with/slashes currently degrades into model=undefined and an Unexpected server error. A clearer model not found error would make diagnosis easier.
4. When Copilot /models returns picker-enabled IDs containing slashes, OpenCode should be able to register and select those IDs without requiring manual opencode.json entries.

[rekram1-node]

GitHub Copilot model discovery should use the same headers official Copilot clients use, or otherwise preserve account-specific enterprise / Experimental model visibility.

Nah we arent supposed to. They gave us specific guidelines and we follow them.

[KTY810]

What I'm trying to say is that my account has the necessary permissions, and I can see the corresponding model on GitHub Copilot and use it there, but by default, it isn't detected on OpenCode.

[KTY810]

This is likely an issue with GitHub Copilot rather than OpenCode.