← ClaudeAtlas

model-pickerlisted

Use for choosing a Claude model and effort level for a task, an agent or a whole project, from Haiku 5.5 up to Fable 5.2. Use it whenever someone asks which model or effort to use, whether Fable, Opus, Sonnet or Haiku is worth it for some work, how to spend fewer tokens or stay inside weekly plan limits, how to set models and effort for Claude Code subagents and agent files, or what a new Claude model release changes. Also use it before handing work to subagents or planning a multi-agent build, even when nobody asked about models. Load it before answering from memory. The lineup, prices and benchmark results change every few weeks and the model's memory of them is dated.
Pupfish-LLC/model-picker · ★ 1 · AI & Automation · score 74
Install: claude install-skill Pupfish-LLC/model-picker
# Model picker Pick the Claude model and effort level that fits the work, the stakes and the way the person pays. The aim is the cheapest pick that still gets the work right the first time. A pick that is too large wastes money or plan allowance. A pick that is too small costs more in the end, because failed attempts and rework are not free. **Last reviewed: 2026-11-12.** Covers Claude Fable 5.2, Opus 5.5, Sonnet 5.5 and Haiku 5.5. Every score and cost behind this guidance is Anthropic's published figure or a reading of an Anthropic chart, not a measurement on the user's own work. Treat the picks as a well-supported starting point. When a choice is close, say so, and suggest the user measure on their own work. ## Step 1. Check that this guidance is still current Do this quickly and silently unless something is off. - Compare today's date with the review date above. If more than about 60 days have passed, or the user names a Claude model that is not in the lineup below, tell the user in one line that this guidance may be behind. - If a web search tool is available, check https://www.anthropic.com/news for a newer model release before answering, and say what you found. - Do not guess where an unknown model fits from its name. Either apply the method in `references/reading-cost-charts.md` to its published charts, or say that the plugin needs an update. ## Step 2. Work out how the person pays The same task can deserve a different pick depending on what the tokens draw on.