Source: Choosing the right model (Anthropic)

Anthropic's guide to picking a Claude model: criteria, efficiency-first vs capability-first starts, a selection matrix and eval-driven upgrades.

Contents

Summary

Frames model choice as balancing capabilities, speed and cost, with effort as a within-model lever. Offers two starting strategies and recommends building use-case-specific evals before switching models.

Key claims

  • Criteria: capabilities, speed, cost, and effort; tuning effort is often a better lever than switching models.
  • Efficiency-first: start with Haiku 5.5, test, upgrade only for proven gaps. Suits prototyping, tight latency, cost-sensitive, high-volume simple tasks.
  • Capability-first: start with Opus 5.5, optimise prompts, then lower effort or downgrade; move to Fable 5.2 only if evals at xhigh/max still fall short.
  • Selection matrix: Fable 5.2 for hours-long agent sessions and deep research; Opus 5.5 for complex agentic coding and enterprise work; Sonnet 5.5 for everyday coding/agent work; Haiku 5.5 for lowest latency/price and subagent tasks.
  • A good evaluation set on your own prompts and data is called the most important step in deciding to upgrade.
  • Opus 5.5 has breaking changes for integrations that force tool use (tool_choice any/tool), turn thinking off, or use the computer_20251124 tool.
  • Multi-model strategies pair a cheap model with a frontier model (executor+advisor, orchestrator+workers).

Feeds

Model selection

Suggest changes →