Claude models · cheat sheet

Which Claude for which job, and what $1,000 buys on each

Sonnet 5.5 (out Sept 28) scored 70.6% on Terminal-Bench 4.0, ahead of Opus 5.5's 66.4%, at Sonnet prices. Opus is still Anthropic's pick for complex work, Fable for the hardest and longest tasks. Here's the cheat sheet plus how many messages $1,000 buys on each model.

Came from the Reel? You commented MODELS. Here is everything, with one-tap copy.

Before you start: Prices are the official Claude API list prices per million tokens (input / output): Haiku 4.5 $1 / $5 · Sonnet 5.5 $2 / $10 · Opus 5.5 $4 / $20 · Fable 5.1 $10 / $50.

Setup

Simple, fast, high-volume tasks
haiku
Everyday coding and writing
sonnet
Complex reasoning and careful judgment
opus
Your hardest, longest-running, most ambiguous tasks
fable

All 2, step by step

What $1,000 buys

Example message: 3,000 tokens in (your question plus the chat so far) and 500 tokens out, with no caching. Longer chats cost more per message because every message re-sends the conversation.

Haiku 4.5: about $0.0055 a message, roughly 180,000 messages.
Sonnet 5.5: about $0.011 a message, roughly 90,000 messages.
Opus 5.5: about $0.022 a message, roughly 45,000 messages.
Fable 5.1: about $0.055 a message, roughly 18,000 messages, 10x fewer than Haiku.

Tip: Prompt caching cuts the re-sent part to a fraction of the price (0.05x on Opus 5.5, 0.025x on Fable 5.1).

Official docs →

Sonnet 5.5 in numbers

Released September 28, 2026. 70.6% on Terminal-Bench 4.0 (Opus 5.5: 66.4%, Sonnet 5: 10.3%). First Sonnet to beat Pokémon Red working only from screenshots. Same $2 / $10 price as Sonnet 5.

Tip: Use Sonnet for everyday jobs and save Opus or Fable for the big ones. Your limits go much further.

Official docs →

Tip: In Claude Code, switch with /model sonnet (or opus, haiku, fable). opusplan uses Opus while planning and Sonnet while doing.