GitHub Copilot provider shows zero models: all models return `model_picker_enabled: false`

Summary

On opencode 1.18.15 (Arch package), the github-copilot provider never appears in the model picker. opencode auth login -p github-copilot works (auth succeeds), but opencode models github-copilot returns "Provider not found", and /models in the TUI shows no Copilot models.

Root cause

In packages/opencode/src/plugin/github-copilot/copilot.ts, the models() hook returns only models where result.pickerEnabled.has(model.api.id):

return CopilotModels.get(...)
  .then((result) => {
    models = result.models
    return Object.fromEntries(
      Object.entries(result.models).filter(([, model]) => result.pickerEnabled.has(model.api.id)),
    )
  })

pickerEnabled (built in models.ts:257) is the set of models with model_picker_enabled: true in the API response:

pickerEnabled: new Set([...remote].filter(([, item]) => item.model_picker_enabled).map(([id]) => id)),

GitHub's GET https://api.githubcopilot.com/models currently returns model_picker_enabled: false for every single model — even ones that are policy.state: "enabled" and fully usable. So pickerEnabled is always empty, the filter drops everything, and the provider surfaces zero models.

Evidence

Using a valid, freshly-authenticated Copilot OAuth token (device flow, client_id=Ov23li8tweQw6odWQebz):

HTTP 200, total models: 47
policy states:  disabled: 18 | none: 23 (internal: copilot-search-*, exec-agent-*) | enabled: 6
model_picker_enabled=true: 0

The 6 policy.state: "enabled" models (gpt-4o, gpt-5-mini, claude-haiku-4.5, gpt-4.1, mai-code-1-flash, mai-code-1.1-flash) pass the existing usable() check (models.ts:207), and chat completions against them work fine:

POST https://api.githubcopilot.com/chat/completions → 200, "Hi there! ..."

So the account/subscription is fine — the filter is the problem. Note: exchanging the OAuth token for a Copilot session token via api.github.com/copilot_internal/v2/token and using that against /models returns 403, so the session-token route is not a workaround either.

Suggested fix

Filter by the already-existing usable() predicate (policy.state !== "disabled" + has max_output_tokens/max_prompt_tokens + has tool_calls) instead of, or in addition to, model_picker_enabled. Either:

• Return result.models (already filtered by usable() inside CopilotModels.get) without the pickerEnabled filter, or
• Union pickerEnabled with the usable set.

model_picker_enabled appears to be a deprecated/no-longer-reliable signal from GitHub's API.

Workaround

A local plugin overriding the provider's models() hook to return all usable models (same usable() criteria, no picker filter) restores all 25 available Copilot models:

// ~/.config/opencode/plugins/copilot-picker-fix.js
export const CopilotPickerFix = async () => ({
  provider: {
    id: "github-copilot",
    async models(provider, ctx) {
      if (ctx.auth?.type !== "oauth") return provider.models
      const base = ctx.auth.enterpriseUrl
        ? `https://copilot-api.${ctx.auth.enterpriseUrl.replace(/^https?:\/\//, "").replace(/\/$/, "")}`
        : "https://api.githubcopilot.com"
      const r = await fetch(`${base}/models`, {
        headers: { Authorization: `Bearer ${ctx.auth.refresh}`, "User-Agent": "opencode/1.18.15", "X-GitHub-Api-Version": "2026-06-01" },
        signal: AbortSignal.timeout(5000),
      })
      if (!r.ok) return provider.models
      const j = await r.json()
      const models = {}
      for (const m of j.data || []) {
        if (m.policy?.state === "disabled") continue
        const lim = m.capabilities?.limits
        if (!lim?.max_output_tokens || !lim?.max_prompt_tokens) continue
        if (m.capabilities?.supports?.tool_calls === undefined) continue
        const isMsg = m.supported_endpoints?.includes("/v1/messages")
        models[m.id] = {
          id: m.id, providerID: "github-copilot",
          api: { id: m.id, url: isMsg ? `${base}/v1` : base, npm: isMsg ? "@ai-sdk/anthropic" : "@ai-sdk/github-copilot" },
          status: "active",
          limit: { context: lim.max_context_window_tokens ?? lim.max_prompt_tokens, input: lim.max_prompt_tokens, output: lim.max_output_tokens },
          capabilities: { temperature: true, reasoning: true, attachment: true, toolcall: m.capabilities.supports.tool_calls, input: { text: true, audio: false, image: !!m.capabilities.supports.vision, video: false, pdf: false }, output: { text: true, audio: false, image: false, video: false, pdf: false }, interleaved: false },
          family: m.capabilities.family, name: m.name, cost: { input: 0, output: 0, cache: { read: 0, write: 0 } }, options: {}, headers: {},
          release_date: m.version.startsWith(`${m.id}-`) ? m.version.slice(m.id.length + 1) : m.version,
        }
      }
      return models
    },
  },
})

Environment

• opencode 1.18.15-1 (Arch Linux package)
• GitHub Copilot subscription: active (chat completions return 200)
• OAuth apps tested: opencode's own (Ov23li8tweQw6odWQebz) and the VSCode Copilot app (Iv1.b507a08c87ecfe98) — same result either way