Claude Sonnet 5.5 vs Opus 5.5: The Cheat Sheet for Which Claude to Use on Every Task
The new Sonnet beat Opus on a coding test the day it came out, at half the price per token. Here is when to use each one, so the work gets done well and your usage limit lasts.
Anthropic released Claude Sonnet 5.5 on September 28. On Terminal-Bench 4.0, a test of real command-line coding work, it scored 70.6%, ahead of Claude Opus 5.5 at 66.4% and far ahead of the last Sonnet at 10.3%. It is also the first Sonnet to beat Pokémon Red working only from screenshots of the game.
It costs half as much per token as Opus, writes over 30% faster than Sonnet 5 and uses fewer tokens per task. But Opus is still ahead on several other tests, and Anthropic still says Opus is the one for complex work. So the question isn't which model is better. It's which one to pick for each job.
What Anthropic shipped
| Test (Anthropic's numbers) | Sonnet 5.5 | Opus 5.5 | Sonnet 5 |
|---|---|---|---|
| Terminal-Bench 4.0 (agentic coding) | 70.6% | 66.4% | 10.3% |
| FrontierCode 1.1 (hard coding) | 46.2% (max effort) | 54.4% | 42.4% |
| CursorBench 4.0 (coding in an editor) | 55.5% | 57.8% | 34.1% |
| GDPval-AA v2.1 (knowledge work) | 1844 | 1846 | 1449 |
| OSWorld 2.1 (using a computer) | 80.1% | 81.8% | 57.0% |
| Humanity's Last Exam (with tools) | 64.5% | 67.7% | 54.9% |
| Price per 1M tokens (in / out) | $2 / $10 | $4 / $20 | $2 / $10 |
Read the table as two groups. On everyday work (command-line coding, knowledge work, using a computer) Sonnet 5.5 matches or beats Opus. On the hardest problems (FrontierCode, Humanity's Last Exam) Opus keeps a clear lead. Anthropic's own advice follows the same line: Sonnet 5.5 is strongest at well-scoped everyday tasks, fixing bugs and making polished documents, slides and spreadsheets; Opus 5.5 is built for complex work that needs careful judgment.
Keep reading
Free, for an email.
Unlocks 4 more sections, the document version, and every other guide on the site, for free. Enter your email once to keep reading.
19,000+ follow where these guides come from.No spam. Unsubscribe in one click.