Claude Fable vs Opus vs Sonnet vs Haiku: which Claude model should you use?
Anthropic now has four Claude tiers. What Fable 5.1, Opus 5.5, Sonnet 5.5 and Haiku 5.5 are each for, what they cost, and the one most people should pick.
We may earn a commission if you sign up through links on this page. It costs you nothing and never decides what we recommend. Learn more
Anthropic’s Claude used to come in three sizes. Since September 2026 there are four, and the new top tier, Fable, sits above Opus. Here’s what each model is for and how to choose without overpaying.
Quick answer: for most work, use Opus 5.5 (or Sonnet 5.5 if you want it faster and cheaper). Use Fable 5.1 for long, hard jobs. Use Haiku 5.5 for quick, high-volume tasks.
The four Claude models
| Model | Released | Best for | API price (in / out, per 1M tokens) | Context |
|---|---|---|---|---|
| Fable 5.1 | Sep 1, 2026 | Long agent runs, deep research, spreadsheets and slides | $10 / $50 | 1M tokens |
| Opus 5.5 | Sep 22, 2026 | The most capable everyday model | $4 / $20 | 1M tokens |
| Sonnet 5.5 | Sep 28, 2026 | Balanced quality and speed | $2 / $10 | 1M tokens |
| Haiku 5.5 | Oct 7, 2026 | Fast, cheap, high volume | $0.10 / $0.50* | 1M tokens |
*Haiku prompts over 100K tokens cost $0.50 / $2.50. Prices from Anthropic’s model pages, checked October 28, 2026.
Fable 5.1: for the hardest jobs
Fable is Anthropic’s top tier. Fable 5.1 launched on September 1, 2026, with a 1 million token context window and up to 128K tokens of output. Anthropic says its biggest gains are in long-running coding agents, multi-step research, and work on documents, spreadsheets and slides.
It’s expensive, $10 / $50 in the API, though cheaper cached input cuts typical bills by about 25% compared with Fable 5. In the Claude apps, Max plans include it; Pro users pay with usage credits. Its twin, Mythos 5.1, is the same model with looser safeguards for vetted security and life-sciences teams only.
Opus 5.5: the default for serious work
Opus 5.5 is the most capable everyday Claude, and it got cheaper: $4 / $20, down from $5 / $25 for Opus 5. Anthropic’s own documentation suggests starting most workloads with Opus 5.5 and moving up to Fable only when you need to.
Sonnet 5.5: the best value
Sonnet 5.5 gives you most of Opus’s quality for half the API price. For writing, summaries and everyday coding it’s often all you need, and it’s faster.
Haiku 5.5: fast and cheap
Haiku 5.5 costs 10 cents per million input tokens. Use it for sorting, tagging, quick replies and first-pass summaries, anything you run hundreds of times a day.
Which should you use?
| If you want to… | Use |
|---|---|
| Write, summarize or code day to day | Sonnet 5.5 |
| Get the best answer on important work | Opus 5.5 |
| Run a long agent or research task for an hour | Fable 5.1 |
| Process thousands of small items cheaply | Haiku 5.5 |
In the Claude app, the plan decides much of this: see is Claude free? for what each plan includes, and our Claude review for the full picture. Comparing across companies? Read GPT-6 models explained.
Tools in this post
Claude
by Anthropic
Assistant known for long documents, careful writing and coding help.
Free; Pro $20/mo ($17/mo yearly); Max from $100/mo
Privacy & risks
- Trains on your data
- You choose
- Minimum age
- 18+
- Free, Pro and Max users choose whether their chats can be used for training. Team, Enterprise and API data is not used for training.
Claude Code
by Anthropic
Coding agent that works across your whole codebase from the terminal or editor.
Included with Claude Pro and Max; or pay per use via the API
Privacy & risks
- Trains on your data
- You choose
- The most-used coding agent in the Stack Overflow 2026 survey (66% of developers who use agents).
- Consumer plans let you choose whether code is used for training; Team, Enterprise and API use is not trained on.
- Security flaws found in 2026 (including CVE-2026-24887 and CVE-2026-35021) were fixed in updates. Keep it up to date.
Questions people ask
Is Claude Fable better than Opus?
Fable 5.1 is Anthropic's most powerful generally available model and is aimed at long, hard tasks. It also costs more than twice as much as Opus 5.5 in the API. Anthropic's own docs suggest starting most workloads with Opus 5.5.
What's the difference between Claude Fable and Claude Mythos?
They are the same model with different safeguards. Fable 5.1 is available to everyone; Mythos 5.1 is only for vetted cybersecurity and life-sciences teams.
Which Claude model is cheapest?
Haiku 5.5, at $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 100K tokens.
