AI Tools
What does Claude Opus 5.5 cost per million tokens?
Claude Opus pricing is $4 per million input tokens and $20 per million output tokens, half what the new Fable 5.2 charges at $8 and $40. Anthropic still calls Opus 5.5 its strongest daily driver, which keeps the cheaper model the sensible place to start.
What matters
- Opus 5.5 is $4 per MTok in and $20 out, exactly half Fable 5.2's $8 and $40, per Anthropic's pricing page read on 12 November 2026.
- A task using 1M input and 200K output tokens bills at $8.00 on Opus 5.5, $16.00 on Fable 5.2, $4.00 on Sonnet 5.5 and $0.20 on Haiku 5.5.
- Cache writes cost $5 per MTok on Opus against $10 on Fable, so long reused system prompts favour Opus before any output is generated.
- Pro at $20 a month includes Opus but only usage credits for Fable; Max at $100 includes Fable up to half the weekly limit.
Claude Opus 5.5 pricing, verified 12 November 2026
Opus 5.5 is billed at $4 per million input tokens and $20 per million output tokens on the Claude Platform. Fable 5.2, released on 9 November 2026, costs $8 and $40 for the same units, so the newer flagship is exactly double on both sides of the meter.
Prompt caching follows the same shape with one difference worth attention: a cache write is $5 per MTok on Opus against $10 on Fable 5.2, while a cache read costs $0.20 on both. Sonnet 5.5 runs at $2 and $10. Haiku 5.5 runs at $0.10 and $0.50, with a fivefold step up once a prompt passes 100K tokens.
Anthropic's launch post is blunt about where to begin: for most coding and knowledge tasks the recommendation is to start with Opus 5.5 and then move up to Fable or Mythos, or down to Sonnet or Haiku, depending on the work. That single sentence reframes the price list. The most expensive model is no longer the default purchase.
What a task actually costs on each model
Per-token prices compare cleanly and read badly. Put 1 million input tokens and 200,000 output tokens through each model and the bill is:
- Fable 5.2: $16.00
- Opus 5.5: $8.00
- Sonnet 5.5: $4.00
- Haiku 5.5: $0.20
Opus and Fable sit 2x apart. Opus and Haiku sit 40x apart on the same workload. Output tokens drive that gap: they cost five times the input rate on every model in the family, so a task that answers at length pays the multiplier on the side of the meter that hurts.
For a support agent or a coding loop that reads a large repository and writes patches, the input side dominates early and the output side dominates once responses get long. Model the expected split before choosing, because the ranking of these four models changes with it.
Cache writes are where Opus pulls further ahead
Agentic work resends the same system prompt and tool definitions on every call. Write 200K tokens into the cache once and Opus charges $1.00 for the write; Fable 5.2 charges $2.00. Read that cache ten times and the totals land at $1.40 on Opus and $2.40 on Fable, against $8.00 and $16.00 if the prompt is resent in full each time.
Cache reads are priced identically at $0.20 per MTok on both models, so the write is the only part of the cache story that separates them. Long, stable contexts therefore tilt the decision toward Opus before a single output token is generated, which is the opposite of what the headline price comparison suggests for short one-off prompts.
Why Fable 5.2 can still cost less per finished task
Token price is not task price. Anthropic reports that on Chartography, a chart-reading evaluation, Fable 5.2 at Medium effort scores 66.4% against Opus 5.5 at Max effort at 64.4%, for less than one-tenth of the cost per task. On Perplexity's WANDR data-collection benchmark, Fable 5.2 at Medium effort matches Opus 5.5's top score for about 40% of the cost per task.
Effort level explains the arithmetic. A model run at Max effort spends far more tokens reasoning, so a model with a higher per-token price at a lower effort setting can finish the same task for less. Both figures come from the vendor's own launch post and describe its internal benchmark runs, not an independent test, which is a reason to treat them as directional rather than as a budget forecast.
Fast mode cuts latency rather than cost. For Opus 5.5 and Fable 5.2 it doubles the token price ($8 and $40, or $16 and $80) to generate up to 2.5x faster, and Anthropic notes Fable 5.2 already produces output up to 30% faster than Fable 5.1 at standard speed.
The subscription route: Opus is in the $20 plan, Fable is not
Most buyers meet these prices through a Claude subscription instead of an API key, and the plan matrix changes which model is affordable. Anthropic's plan comparison lists Opus on Pro at $20 a month ($17 with annual billing), alongside Sonnet and Haiku. Fable is the exception: on Pro it appears as usage credits, and on Max ($100) and Ultra ($200) it is included up to half the weekly limit.
Usage limits scale on the same axis. Pro carries 1x the 5-hour and weekly allowance, Max carries 5x, and Ultra removes the 5-hour cap while giving 10x the weekly usage. Max includes $100 a month in API credits and Ultra $200, which is the only point where the subscription and the per-token meter meet.
The practical reading: $20 buys Opus 5.5 at full plan access. Fable 5.2 at volume starts at $100, because Pro meters it through credits rather than including it. Anyone comparing Claude Opus pricing against Fable subscriptions is comparing two different purchase mechanisms, not two line items.
Multipliers that move the Opus bill
Three modifiers apply across the price list. Batch processing halves token prices, so Opus drops to $2 and $10 and Fable to $4 and $20. US-only inference adds 10% on input and output tokens for workloads that must stay in the United States. Fast mode doubles them.
The 100K threshold on Haiku 5.5 is the other cliff. Above it, input goes from $0.10 to $0.50 per MTok and output from $0.50 to $2.50, a fivefold jump that turns a cheap router into a mid-priced model the moment a prompt carries a long document. Opus and Fable do not publish that split in the same table, so Haiku is the model where prompt length most changes the answer.
Which Claude model to default to
- Start on Opus 5.5 for coding, agentic loops and long-context work. It is half Fable's price and the model Anthropic names as its daily driver.
- Escalate to Fable 5.2 when a task fails at Max effort on Opus, or when chart, vision or open-ended research work needs the extra step up.
- Drop to Sonnet 5.5 for high-volume classification, extraction and first drafts where $2 and $10 fit the budget.
- Use Haiku 5.5 for routing, tagging and prompts under 100K tokens, where input costs $0.10 per MTok.
- Ignore Mythos 5.2 unless you are in a vetted program: it is the same underlying model as Fable, with access gated behind Anthropic's cyber and life-sciences verification programs.
The test is not which model is best. It is whether the cheaper model fails the task, and Opus 5.5 fails less often than its price suggests.
At a glance
| Model | Input | Output | Cache write | Cache read | Notes |
|---|---|---|---|---|---|
| Fable 5.2 | $8 | $40 | $10 | $0.20 | Fast mode $16/$80; up to 30% faster output than 5.1 |
| Opus 5.5 | $4 | $20 | $5 | $0.20 | Anthropic's recommended default; fast mode $8/$40 |
| Sonnet 5.5 | $2 | $10 | $2.50 | $0.10 | High-volume coding and agent work |
| Haiku 5.5 | $0.10 | $0.50 | $0.125 | $0.01 | 5x higher above 100K tokens: $0.50 / $2.50 |
FAQ
Is Claude Opus 5.5 cheaper than Claude Fable 5.2?
Yes, by half. Opus 5.5 costs $4 per million input tokens and $20 per million output tokens, while Fable 5.2 costs $8 and $40, per Anthropic's pricing page read on 12 November 2026.
Does Claude Opus 5.5 have a fast mode?
It does. Fast mode doubles the token price to $8 and $40 in exchange for up to 2.5x faster generation, and Anthropic lists the same option for Fable 5.2 at $16 and $80.
Is Claude Opus included in the $20 Pro plan?
Yes. Pro at $20 a month includes Opus, Sonnet and Haiku, with 1x usage limits. Fable is the exception: Pro shows it as usage credits, and full inclusion starts on Max at $100 up to half the weekly limit.
Related reading
Claude API pricing by model, Claude Fable 5.2 pricing, Claude Pro vs Max, Claude tool profile