Model Scorecard

OpenCode TUI plugin that shows model benchmark score, token cost, and token efficiency directly on the home screen so you can choose models with local quality and cost context.

The status line stays visible while you work:

model score: 77 | token cost (in / out): $5.00 / $25.01 per 1M | token efficiency: 0.08

Install

Install globally into your OpenCode TUI config:

opencode plugin opencode-model-scorecard -g

OpenCode installs npm plugins automatically on startup and caches packages under its own cache directory.

For a project-local install, add the TUI plugin to tui.json or .opencode/tui.json:

{
  "$schema": "https://opencode.ai/tui.json",
  "plugin": ["opencode-model-scorecard"]
}

Open OpenCode and use the Show model benchmark details command to inspect the Terminal-Bench rows, token prices, and token-efficiency calculation for the currently selected model.

Why This Exists

Model choice in OpenCode is usually a tradeoff between coding-agent quality, input/output price, and how many output tokens a model tends to spend to solve tasks. Those signals are scattered across benchmark pages, provider catalogs, and local usage. Model Scorecard puts the most useful comparison line in the place where you pick and use models.

Data Sources

• Model score: Terminal-Bench 2.1 when available, Terminal-Bench 2.0 as a fallback, then Artificial Analysis Terminal-Bench v2.1 when official rows are missing.
• Token cost: OpenCode Data model catalog.
• Token efficiency: Artificial Analysis Terminal-Bench v2.1 score and output-token data when available, with OpenCode session-cost rows retained as a real observed fallback.

All data is bundled into the package at release time and resolved locally at runtime. The plugin does not call a backend or fetch benchmark data while you use OpenCode.

Known Limits

• Models without a Terminal-Bench or Artificial Analysis match show n/a for benchmark-derived fields.
• Benchmark rows are useful comparison signals, not absolute model rankings for every codebase.
• Prices and benchmark coverage update only when maintainers publish a refreshed npm release.
• Provider aliases can differ. Please open a missing-model issue when a model should resolve but does not.

Score Refreshes

Maintainers can refresh the hardcoded benchmark registry before publishing a new npm release:

npm run update:scores
npm test
npm run pack:dry-run
npm version patch
npm publish

The refresh command uses Terminal-Bench 2.1, Terminal-Bench 2.0, Artificial Analysis Terminal-Bench v2.1, and OpenCode Data, then rewrites only the marked data arrays in .opencode/plugins/model-score-data.mjs. End users do not need a backend or a network call at runtime.

Adoption

See docs/adoption.md for the discovery plan, validation checks, listing copy, and launch assets used to make this plugin easier for OpenCode users to find and evaluate.