[BUG]: Custom openai-compatible provider model limit overrides ignored for context percentage display #29555

Description

User-configured limit.context for a custom @ai-sdk/openai-compatible provider is ignored when computing the context usage percentage in the TUI.

Config:

{
  "provider": {
    "custom": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "custom (llama-server)",
      "options": {
        "baseURL": "http://<local-host>/llama-3/v1"
      },
      "models": {
        "qwen3.6-27b": {
          "name": "Qwen3.6 27B",
          "limit": { "context": 128000, "output": 65536 }
        }
      }
    }
  }
}

Expected: With ~95K tokens used and context: 128000, the UI should show ~74%.

Actual: The UI shows "95.1K / 47%". The percentage implies a context of ~202K (95.1K / 0.47 ≈ 202K), which is not what's configured. This suggests the context limit is being resolved from the models.dev database or the llama-server's /v1/models response, and the user's limit override is being ignored for display purposes.

OpenCode version

1.15.11

Steps to reproduce

1. Configure a custom provider with @ai-sdk/openai-compatible pointing to a llama-server instance.
2. Define explicit limit.context and limit.output for a model in the provider config.
3. Start a session and accumulate tokens.
4. Observe the context percentage in the TUI — it does not match the configured limit.context.

Operating System

Linux

Terminal

Alacritty

Related issues

• GitHub Copilot plugin ignores user-configured model limit overrides #21564 — Copilot plugin ignores user-configured model limit overrides (same class of bug, different plugin)
• [BUG]: Context percentage indicator doesn't reflect usable context proportion #28057 — Context percentage indicator doesn't reflect usable context proportion

Note: This issue was filed by an AI assistant (Qwen3.6, running inside opencode) at the user's request, after investigating the config, schema, and existing issue tracker.