Anthropic’s Claude 5.5 family has three tiers. This guide uses the prices and positioning Anthropic publishes; check its pages before you budget, since they change.
The three models at a glance
| Model | Released | Input / output per million tokens (Anthropic’s list) | Anthropic positions it for |
|---|---|---|---|
| Opus 5.5 | Sept 22, 2026 | $4 / $20 | Complex, long-running work; roughly Fable 5.1 level on most tasks, says Anthropic |
| Sonnet 5.5 | Sept 28, 2026 | $2 / $10 | Everyday tasks, bug fixes, polished documents |
| Haiku 5.5 | Oct 7, 2026 | $0.10 / $0.50 (prompts up to 100,000 tokens) | High-volume, cost-sensitive work and subagents |
A simple way to choose
- Start with Sonnet 5.5 for writing, analysis and normal coding help.
- Move up to Opus 5.5 when the task is large or multi-step and a wrong answer is costly, such as auditing a big codebase.
- Drop to Haiku 5.5 for summaries, classification, tagging and live support, where speed and price matter more. It has an adjustable effort setting, so you can trade cost for intelligence.
- Test on your own examples. Run the same prompt through two tiers and compare. Our prompt workflow guide shows how.
A prompt to test all three (our own)
You are reviewing a short customer email. Summarise it in two sentences, label the tone as calm, frustrated or urgent, and draft a reply of under 80 words. Then list one thing you were unsure about.
If Haiku and Sonnet give the same useful answer, use the cheaper one at scale.
Good to know
- Opus 5.5 cannot have thinking switched off, per Anthropic.
- Some biology and cybersecurity requests are handled under extra safeguards.
- Tokenizers differ between versions, so token counts for the same text can change.
For the wider picture see the Claude tool page and ChatGPT vs Claude. More model news is in the model tracker.
Disclosure: Anthropic’s Claude is one of the tools used to help build this site. We cover Claude models on the same basis as every other maker.