Anthropic Claude family
Anthropic Claude Models and Token Costs: Fable, Opus, Sonnet and Haiku
The current Claude model families, how each tier is meant to be used, what each version costs per million tokens, and which older versions are being retired.
Prices and versions: figures are published list prices in US dollars per 1 million tokens, read from each provider’s own pages on 18 November 2026. Models, prices and promotions change often, so confirm on the provider’s page before you budget.
The Claude tiers
- Fable. Anthropic positions Fable for demanding reasoning and long-horizon agentic work, or when evaluations on Opus 5.5 at higher effort still fall short.
- Opus. Aimed at long-running agentic coding and knowledge work, and the model Anthropic points most workloads towards.
- Sonnet. Anthropic describes it as the best combination of speed and intelligence, and it is the usual production default.
- Haiku. For high-volume, latency-sensitive tasks such as classification, extraction and routing.
- Mythos. Listed alongside Fable with the same prices at each version. The pages we read give no tier description, so confirm its availability and access conditions with Anthropic.
Token prices by model
List prices in US dollars per 1 million tokens for the first-party API. Cache read is the discounted price for reused prompt content.
| Model | API id | Input | Output | Cache read | Context / max out |
|---|---|---|---|---|---|
| Fable 5.2 | claude-fable-5-2 | $8 | $40 | Check page | 1M / 128K |
| Fable 5.1 | claude-fable-5-1 | $10 | $50 | $0.25 | 1M / 128K |
| Opus 5.5 | claude-opus-5-5 | $4 | $20 | $0.20 | 1M / 128K |
| Opus 5 | claude-opus-5 | $5 | $25 | $0.50* | 1M / 128K |
| Sonnet 5.5 | claude-sonnet-5-5 | $2 | $10 | $0.10 | 1M / 128K |
| Sonnet 4.6 | claude-sonnet-4-6 | $3 | $15 | $0.30* | 1M / 128K |
| Haiku 5.5 | claude-haiku-5-5 | $0.10 (prompts up to 100K) | $0.50 | $0.01 | 1M / 128K |
| Haiku 4.5 | claude-haiku-4-5-20251001 | $1 | $5 | $0.10* | 200K / 64K |
* Calculated from the documented default cache-read rate of one tenth of the input price. Fable 5.2 appears on Anthropic’s models overview at $8 and $40, while the pricing table we read still labels its top row Fable 5.1 at $10 and $50, so verify the current row before budgeting. Fable 5 and Mythos 5 sit at $10 and $50. Anthropic lists a reliable knowledge cutoff of June 2026 for Fable 5.2, Opus 5.5, Sonnet 5.5 and Haiku 5.5.
What a monthly bill looks like
A worked example: 100 million input tokens and 20 million output tokens a month, no caching or batching. Real bills move with your cache hit rate, output length and any long prompts.
| Model | Monthly cost: 100M input + 20M output tokens |
|---|---|
| Fable 5.1 | $2,000 |
| Fable 5.2 | $1,600 |
| Opus 5.5 | $800 |
| Sonnet 4.6 | $600 |
| Sonnet 5.5 | $400 |
| Haiku 5.5 (prompts up to 100K) | $20 |
What changes the bill besides the model
- Prompt caching. Writes to the cache cost more than normal input (a premium for the five-minute and one-hour options), and reads cost a small fraction of it. Reads are 0.1 times input by default, and lower on some newer models.
- Batch API. A 50 percent discount on input and output for work that can wait.
- Long prompts. Haiku 5.5 is the one model with a long-prompt surcharge: prompts over 100,000 tokens move to $0.50 in and $2.50 out. Other current models carry the full 1M window at standard rates.
- Fast mode. A research-preview, API-only option that charges a higher rate for faster output on selected Opus models.
- US-only inference. Choosing US data residency adds a 1.1 times multiplier on the newer models.
- Tokenisers differ. Different model generations and different vendors count tokens differently, so price per token is not price per task.
Version lifecycle
Anthropic publishes deprecation and retirement dates. Pin full model ids and track the schedule, because pinned old versions do get retired.
| Model | Status on the model pages |
|---|---|
| Fable 5.2, Mythos 5.2 | Active, not retiring sooner than 9 Nov 2027 |
| Opus 5.5, Sonnet 5.5, Haiku 5.5 | Active, not retiring sooner than Sep to Oct 2027 |
| Opus 4.5 | Active, not retiring sooner than 24 Nov 2026 |
| Sonnet 4.5 | Deprecated, retires 30 Nov 2026; replacement is Sonnet 5.5 |
| Haiku 4.5 | Active, not retiring sooner than 15 Oct 2026 |
| Opus 4.1, Opus 4, Sonnet 4 | Retired in June and August 2026 |
| Haiku 3.5, Haiku 3 | Retired in February and April 2026 |
| Mythos Preview | Deprecated 9 June 2026, retirement date to be announced |
How to choose a tier
- Start with Sonnet. The default for production assistants, coding help and most agent steps.
- Use Haiku for volume. Routing, extraction and sub-agents, where latency and price matter most.
- Escalate to Opus or Fable. Long-running coding agents and hard reasoning, ideally selected by a router rather than hard-coded.
- Keep a fallback. Retirements and rate limits happen, so test a second model before you need it.
Where this meets infrastructure
Claude models are served through Anthropic’s own API and through major cloud platforms, and many teams reach them through a gateway such as OpenRouter. We help decide when API access is right and when hosted or self-run open models make more sense, and we build the surrounding agents and security controls. See hosted AI models for dedicated capacity.
Where this fits in the stack
Common questions
Which Claude model should I use?
Most teams start with Sonnet, use Haiku for high-volume simple tasks, and escalate to Opus or Fable for long-running agentic work or hard reasoning. Test on your own data and compare cost per successful task.
How much does Claude cost per million tokens?
On the pages we read, current input prices run from $0.10 (Haiku 5.5) through $2 (Sonnet 5.5) and $4 (Opus 5.5) to $8 to $10 (Fable), with output at five times input.
How do I lower Claude API costs?
Cache stable prompt prefixes, use the Batch API for non-urgent work, pick the smallest tier that passes your evaluations, keep prompts short, and avoid Haiku 5.5 prompts above 100,000 tokens if cost matters.
Are old Claude versions retired?
Yes. Anthropic lists deprecation and retirement dates for each model, and several older Opus, Sonnet and Haiku versions have already been retired or are scheduled to be.
Sources and dating
Source: Anthropic documentation (pricing, models overview, model deprecations and model ids pages), read 18 November 2026. Cache and batch figures for some rows come from the documented multipliers, and Mythos pricing is as listed on the overview. Some figures are vendor claims or preliminary and some differ between sources, so we confirm against the provider pages before we size or budget a build. Anthropic, Claude, Fable, Mythos, Opus, Sonnet and Haiku are trademarks of their owners. Embedica is independent and is not affiliated with or endorsed by any of the companies named.
Next steps
Related: LLM models and token costs · OpenAI models · Open-source AI models · AI application layer
Need help choosing and budgeting for Claude models?
Model and cost review
