Full reference for model selection. See the Agent Roster for the per-agent quick summary.
| Need | Provider | Pool |
|---|
| Primary reasoning | Opus 4.6 | Anthropic MAX (subscription) |
| Sub-agent coding | Codex (GPT-5.3) | ChatGPT Plus (unlimited) |
| Bulk/parallel | GPT-4o | Copilot Pro (unlimited) |
| Large doc analysis | Gemini 3.1 Pro | Google OAuth (free) |
| Fast lightweight | Gemini Flash | Google OAuth (free) |
| Secondary Claude | Copilot Sonnet 4.6 | Copilot Pro (300/mo) |
| Premium secondary | Copilot Opus 4.6 | Copilot Pro (100/mo) |
| Tier | Complexity | Examples | Route To |
|---|
| T1 — Routine | Single-file, clear spec | Bug fix, config edit, simple script, CSS tweak | Codex CLI → Gemini CLI |
| T2 — Standard | Multi-file, well-defined | New feature (clear spec), module refactor | Codex CLI → Gemini CLI → Sonnet |
| T3 — Complex | Architectural, ambiguous | Design system work, cross-service coordination | Opus 4.6 |
| T4 — Critical | Security, QA, production | Security audit, code review | Opus 4.6 |
Escalation rule: T1 fails → retry with other free tool → T2. T2 free tools fail → Sonnet. Sonnet fails or task is ambiguous → Opus.
| Model | Provider String | Cost | Best For |
|---|
| Opus 4.6 | anthropic/claude-opus-4-6 | Max plan | T3/T4 coding, orchestration, QA |
| Sonnet 4.6 | github-copilot/claude-sonnet-4.6 | Copilot Pro | T2 fallback, writing, frontend |
| Codex (GPT-5.3) | openai-codex/gpt-5.3-codex | Free (Plus plan) | T1/T2 coding, config/infra |
| Gemini 3.1 Pro | google-gemini-cli/gemini-3.1-pro-preview | Google OAuth | Research, data processing |
| Gemini Flash | google-antigravity/gemini-3-flash | Google OAuth | High-volume triage, budget tasks |
| GPT-4o | github-copilot/gpt-4o | Copilot Pro | Bulk/parallel |
- Default coding to unlimited/free tools first.
- Do NOT use OpenRouter for sub-agents — credits are limited.
# Codex CLI — free via ChatGPT Plus. Terminal-Bench 77.3%. Best for T1/T2.
codex --model gpt-5.3-codex --approval-mode full-auto --quiet "task"
# Gemini CLI — free. 1M+ context. Good for T1/T2 + research.
gemini -m gemini-3.1-pro -ni --sandbox permissive "task"
# Claude Code CLI — Max plan. Use for T2/T3.
claude -p --dangerously-skip-permissions --model sonnet --max-budget-usd 3 "task. Build. Fix errors. Commit."