benchrPricing breakdown
| Tier | Rate / 1M tokens |
|---|---|
| Standard input | $10.00 |
| Standard output | $50.00 |
| Cached input | $1.00 |
| Context window | 1,000,000 tokens |
| Max output | 128,000 tokens |
| Claude plans | Usage credits after the July 7 launch inclusion |
Availability is wide from day one: the Claude API, Claude Platform on AWS, Amazon Bedrock, Vertex AI, and Microsoft Foundry, all under the id claude-fable-5. Claude Mythos 5, the same model with safeguards lifted, lists at the same $10/$50 but is restricted to approved Project Glasswing organizations, so for budgeting purposes Fable 5 is the Mythos-class price.
How tokenizer changes affect the bill
Fable 5 uses the tokenizer introduced with Opus 4.7. Anthropic's documentation says it produces roughly 30% more tokens than pre-4.7 models for the same text. That affects comparisons with older Claude bills: the same text can face both a different rate and a higher token count.
Opus 4.7 and 4.8 share the tokenizer, so their token-count comparison is more direct. When the baseline is an older model, include Anthropic's approximate 1.3× token-count guidance in the cost scenario and verify it on representative prompts.
Caching math at the $10 tier
The absolute saving from caching grows with the standard rate. A 100K-token system prompt called 1,000 times a day bills $1,000/day at the standard rate. At Fable 5's $1/1M cached-input rate, the repeated portion is about $100/day plus the first-write cost.
Eligibility, writes, expiry, and output still affect the bill. Model your exact mix in the cost calculator.
Refusal and fallback economics
Fable 5 can refuse requests caught by its safety classifiers. Anthropic documents a refusal as HTTP 200 with stop_reason: "refusal"; it does not automatically return an Opus 4.8 answer. A request refused before output is generated is not billed.
You can configure server-side, client-side, or manual fallback. Anthropic says its fallback credit refunds the prompt-cache cost of switching so the cache is not paid twice, while the retry is billed at the serving model's rate. Handle the refusal in code and budget for the fallback model you select.
How to evaluate the price premium
Anthropic's launch table reports the Mythos-class pair at 80.3% on SWE-bench Pro against 69.2% for Opus 4.8. These are provider-reported results, not independent benchr measurements. At the listed rates, a coding agent using 2M input and 400K output tokens daily costs about $40/day on Fable 5 and $20/day on Opus 4.8.
Whether the premium pays back depends on acceptance rate, retries, and engineer time. Test both models on the same representative tasks and include human review in the comparison. The launch coverage provides more capability context.
The launch inclusions have ended
The original June 9–22 inclusion was interrupted by the June suspension. When access returned July 1, Anthropic included Fable 5 for up to 50% of weekly usage limits on eligible paid plans through July 7. Both periods are now historical. Current Claude-plan use requires usage credits, so evaluate it against an explicit credit budget and verify the live plan terms before purchase.
Use-case fit
Consider for: Long-horizon coding, large migrations, research analysis, and vision-heavy agent loops where a measured quality gain offsets the token premium.
Skip if: Your workload is routine and a cheaper tier meets the need; your application cannot handle explicit refusal and fallback responses; or 30-day retention fails your compliance review.