Model Selection: More Models Than You Think

The Model Landscape

• The model landscape is wider than you think — which tool you use determines your options.
• Claude Code — Anthropic models only
  • › Opus ($$$): deep reasoning, architecture, code review (1M context)
  • › Sonnet ($$): everyday coding, PR review, implementation
  • › Haiku ($): fast lookups, file search, simple transforms
  • › Switch with /model or /fast for Sonnet quick-mode
• Copilot CLI & VS Code Chat — multi-provider
  • › Anthropic: Claude Opus 4.6, Sonnet 4.6, Haiku 4.5
  • › OpenAI: GPT-5.4, GPT-5.3-Codex, GPT-5.2, GPT-5.1-Codex, GPT-5 mini
  • › Google: Gemini 2.5 Pro, Gemini 3.1 Pro, Gemini 3 Flash
  • › xAI: Grok Code Fast 1
  • › Fine-tuned: Raptor mini, Goldeneye
  • › Switch mid-session with /model — each model family has strengths
• Copilot agent fallback chains — specify `model: [claude-sonnet-4-6, gpt-5.4]` in agent config. If Claude is down, fall back to GPT automatically. Claude Code doesn't support this.

Tier Wisely, Save 10x

1. 1 The tiering principle — use the cheapest model that can do the job. Our agents use 3 tiers: Opus (code review), Sonnet (execution), Haiku (lookups). This saves 10x vs using Opus for everything.
2. 2 Match model to task
   • › Deep reasoning/architecture → Opus
   • › Everyday coding/implementation → Sonnet
   • › Fast lookups/simple transforms → Haiku
3. 3 In Copilot CLI, run /model to see all available models — try the same prompt with Claude and GPT to compare approaches
4. 4 Review your agent definitions and ensure each uses the cheapest model tier that produces good results