Copilot Model Analyzer

A VS Code extension that recommends the right GitHub Copilot model for your task — helping teams on Copilot Business reduce cost by routing to the most efficient model rather than always defaulting to the most powerful one.

Usage

In Copilot Chat, prefix your prompt with @analyze:

@analyze write integration tests for the auth service
@analyze fix the null pointer error in UserController
@analyze review all our API handlers for security issues
@analyze refactor the payment module to use the new SDK
@analyze design the architecture for a new notifications service

The extension responds with:

• Recommended model — matched to your task type and scope
• Estimated tokens — input + output split
• Estimated cost — in USD, using Copilot Business rates
• Context fit — whether the task fits the model's context window, with an upgrade suggestion if not

How it works

Two-tier classification

1. Rules-based (zero cost) — keyword and scope detection covers ~80% of tasks instantly
2. LM fallback (~$0.0001/call) — ambiguous prompts are classified by gpt-5.4-nano via the VS Code Language Model API (no separate API key required — uses your existing Copilot auth)

Model routing

Supported providers

Model data is sourced from:

• OpenAI — pricing and specs from models-page-data.react JS bundle on the compare page
• Anthropic — Claude Haiku 4.5, Sonnet 4.6, Opus 4.8
• Google — Gemini 2.5 and 3.x series

Copilot Business billing context

• 1 AI credit = $0.01 USD
• Included: 1,900 credits / user / month (pooled org-wide)
• Promotional: 3,000 credits / user / month until Sep 1 2026
• Code completions and next-edit suggestions are unlimited (no credit cost)
• Chat, CLI, cloud agent, and Spaces interactions are metered

Requirements

• VS Code 1.90 or later
• GitHub Copilot Business subscription
• GitHub Copilot Chat extension

Extension settings

Project-level overrides via .github/copilot-advisor.json:

{
  "budget": {
    "warn_per_session_usd": 0.50,
    "hard_cap_tokens": 350000
  },
  "defaults": {
    "analysis": "gpt-5.4-mini",
    "test_generation": "gpt-5.4-mini",
    "refactor": "gpt-5.2",
    "agentic_wide": "gpt-5.4",
    "architecture": "gpt-5.5"
  },
  "override_allowed": true
}

License

MIT