Anthropic shipped four new models between September and October 2026: Fable 5.1, Opus 5.5, Sonnet 5.5, and Haiku 5.5. Together with Opus 4.8 and Sonnet 4.6 still in service, the Claude family now runs six models.
This page lists official pricing, specs, and selection guidance. Prices are from Anthropic's official pricing page as of November 2026. Models marked with an asterisk are older models not on the current pricing table; their prices are historical.
All models at a glance
| Model | Input | Output | Cached read | Context | Released |
|---|---|---|---|---|---|
| Fable 5.1 | $10 | $50 | $0.25 | 1M | Sep 1 |
| Opus 5.5 | $4 | $20 | $0.10 | 1M | Sep 22 |
| Sonnet 5.5 | $2 | $10 | $0.20 | 1M | Sep 28 |
| Haiku 5.5 | $0.10¹ | $0.50¹ | $0.05 | 1M | Oct 7 |
| Opus 4-8* | $5 | $25 | — | — | — |
| Sonnet 4-6* | $3 | $15 | — | — | — |
¹ Per request under 100,000 tokens. Above 100K, input rises to $0.50 and output to $2.50.
² Cache write $0.50, cache read $0.05 (both within the 100K-token limit).
All four new models share 1 million token context windows and 128K maximum output. Cached read pricing was slashed across the board, so long-running conversations and system prompts stored in cache save a lot.
Which one to pick
Prices span a hundred-fold from $0.10 to $10 per million input tokens. Task complexity spans as far.
Haiku 5.5 ($0.10 / $0.50): the budget tier. Batch classification, simple extraction, light rewrites. Sits at the exact same price point as GPT-6 Luna. Watch the 100K-token cliff: do not send long documents through it.
Sonnet 5.5 ($2 / $10): the daily driver. Coding, writing, analysis. Stronger and cheaper than the previous Sonnet 4.6 at $3/$15. "Sonnet vs Opus" is the top autocomplete suggestion for "claude sonnet".
Opus 5.5 ($4 / $20): the previous flagship, now one tier down from Fable 5.1. But cheaper and stronger than the old Opus 4-8 at $5/$25. Still the go-to for complex long-running agent work and deep reasoning.
Fable 5.5 ($10 / $50): the most capable and most expensive. Cached read at $0.25 gives it the best cache economics in the family. Worth it for tasks where getting it wrong costs more than getting it right.
The 2026 price-cut timeline
Anthropic's pricing has been moving one way this year:
- September 1: Fable 5.1 launches, cache read drops to $0.25
- September 22: Opus 5.5 replaces Opus 4-8, price falls from $5/$25 to $4/$20
- September 28: Sonnet 5.5 lands at $2/$10, cheaper than Sonnet 4.6
- October 7: Haiku 5.5 debuts at $0.10/$0.50, a tenth of Haiku 4.5's $1/$5
The biggest beneficiaries are users in the lightweight tier: a few hundred daily queries that cost tens of dollars a month under the old pricing now cost a few. Provided you stay under 100K tokens on Haiku 5.5.
How to connect
All Claude models speak the Anthropic Messages API (/v1/messages) and support OpenAI Chat Completions compatibility. Different request shapes, same authentication and endpoint.
On HiAPI, Opus 5.5 runs at $2.02 input, roughly half the official $4. As Sonnet 5.5, Haiku 5.5, and Fable 5.1 come online, the same key at https://api.hiapi.ai/v1/chat/completions will switch between Claude and GPT models without changing anything else.
See the Claude Code and Codex setup guide.
How cache saves you money
The biggest per-model cost difference is not input or output pricing. It is the cache.
| Model | Cache write | Cache read | Cached read / input ratio |
|---|---|---|---|
| Fable 5.1 | $2.50 | $0.25 | 2.5% |
| Opus 5.5 | $2.00 | $0.10 | 2.5% |
| Sonnet 5.5 | $2.00 | $0.20 | 10% |
| Haiku 5.5 | $0.50 | $0.05 | 50% |
Cache writes are a one-time up-front cost. Cache reads are the repeat savings on every future hit. A long conversation with a shared system prompt pays back the cache write in the first turn. For Fable 5.1 and Opus 5.5, cached reads cost 2.5% of full input, which is practically free.
FAQ
Why does Claude have so many models?
Staggered pricing mapped to capability, the same approach as OpenAI's Astra, Sol, Terra, Luna. How much the task is worth decides the tier.
Haiku 5.5 versus GPT-6 Luna, which wins?
Official prices are identical at $0.10/$0.50. The difference: Luna has no 100K-token price cliff, so long requests are cheaper on Luna.
Can Sonnet 5.5 replace Opus 5.5?
Depends on the workload. Coding and writing are usually fine on Sonnet 5.5. Long-running agents and heavy reasoning still favor Opus 5.5. Run a round on Sonnet first; switch up only when needed.
Why is HiAPI cheaper than the official price?
Opus 5.5 runs at $2.02 input on HiAPI, about half the official $4. Check the live pricing API for the current figure rather than relying on a cached page.
What is a cache write?
Prompt Caching splits cost into a one-time write, which processes the text, and a read, which retrieves it on every subsequent use. Reads are a fraction of writes, as little as 2.5% on the top models.
Bottom line
Claude's pricing has been heading one direction all year. Four models now cover $0.10 to $10 per million input tokens. Sonnet 5.5 is the default, Haiku 5.5 is the budget choice, and Fable 5.1 is for the work where failure is not an option. Cached reads are the habit that pays off fastest.
Sources
- Official pricing: Anthropic pricing page, https://platform.claude.com/docs/en/about-claude/pricing, and model overview, https://platform.claude.com/docs/en/models/overview (verified November 2026)
- Price-cut and rename timeline: Anthropic release notes combined with third-party aggregators
