GitHub Copilot - GoModel

The github-copilot provider calls the GitHub Copilot API with the models your Copilot plan includes (Free, Pro, Pro+, Max, Business, or Enterprise). Usage draws on the plan’s GitHub AI Credits, not a model vendor’s API key. The type is github-copilot, not copilot, so it is not confused with Microsoft Copilot.

Configure

The credential is a GitHub OAuth token (gho_...) for an account with Copilot access:

GITHUB_COPILOT_API_KEY=gho_...

Or in config.yaml:

providers:
  github-copilot:
    type: github-copilot
    api_key: "${GITHUB_COPILOT_API_KEY}"

GoModel sends the token as-is, with no exchange or refresh step.

Getting a token

Any GitHub OAuth token for the account works, for example:

# GitHub CLI
GITHUB_COPILOT_API_KEY=$(gh auth token)

# A Copilot client sign-in, e.g. after `opencode auth login`
GITHUB_COPILOT_API_KEY=$(jq -r '."github-copilot".refresh' ~/.local/share/opencode/auth.json)

Which models answer depends on your Copilot plan, not on the token. A model the plan does not include answers model_not_supported, even when /models lists it.

Business and Enterprise

Individual plans (Free, Pro, Pro+, Max) use the default base URL, https://api.githubcopilot.com. A Business or Enterprise seat may be served from another endpoint; set it when requests fail with 401 or model_not_supported:

GITHUB_COPILOT_BASE_URL=https://api.business.githubcopilot.com

Your account’s endpoint is endpoints.api in the reply of curl -H "Authorization: Bearer $GITHUB_COPILOT_API_KEY" https://api.github.com/copilot_internal/user.

Models

GoModel lists models from Copilot’s /models. Models your plan lists but the account has not enabled (policy disabled) are left out; enable them in your Copilot settings and they appear on the next model refresh. To serve only some of them, filter the list:

GITHUB_COPILOT_MODEL_FILTER_INCLUDE=claude-sonnet-5.5,gpt-5.6-luna,gpt-4.1

GITHUB_COPILOT_MODELS is a fallback list, used only when /models fails or is empty; CONFIGURED_PROVIDER_MODELS_MODE=allowlist makes it the only list and skips /models.

Chat completions and Responses on every model

Copilot serves GPT-5 and later only on /responses, and Claude, Kimi, and older GPT models only on /chat/completions. GoModel reads each model’s endpoints from /models and serves both APIs on every model:

Client calls	Model served on	GoModel
`/v1/chat/completions`	`/chat/completions`	Forwards
`/v1/chat/completions`	`/responses` only	Translates to Responses and back
`/v1/responses`	`/responses`	Forwards
`/v1/responses`	`/chat/completions` only	Translates to chat completions and back

Until GoModel has read a model’s endpoints, for example while /models is unreachable, the model is sent to /chat/completions. A chat request to a Responses-only model is rejected with 400 when it carries a field the translation cannot map exactly (for example logit_bias); send it to /v1/responses instead. Embeddings (text-embedding-3-small) are supported. Copilot only accepts a list as input, so a single string is wrapped into a one-element list.

Cost

Copilot bills tokens at each model’s listed API rate, in AI credits drawn first from the plan’s monthly allowance. GoModel prices the same traffic from its model catalog, so cost is an estimate of the credits used. It does not account for the included allowance or Copilot’s discounts, so it can exceed what GitHub actually charges.

​Configure

​Getting a token

​Business and Enterprise

​Models

​Chat completions and Responses on every model

​Cost