Quick Answer
Claude Fable 5.1 costs $10 per million input tokens and $50 per million output tokens, unchanged from Fable 5, with cache reads cut 75% to $0.25. Claude Mythos 5.1 shares the same rates but is restricted to vetted organizations. Batch processing halves input and output, and GPT-6 Astra now matches the headline price exactly.
Anthropic released Claude Fable 5.1 and Claude Mythos 5.1 on September 1, 2026, and left the headline rates alone. Two days later, OpenAI launched GPT-6 Astra at the identical $10 and $50, though Astra opened through a limited early-access program rather than to everyone. Headline rate is no longer a reason to pick one lab over the other.
The real pricing news hid one line down the rate card: cache reads dropped 75%, to $0.25 per million tokens. Anthropic says that cuts effective costs around 25% on typical workloads and up to roughly 45% on agentic ones, based on four weeks of its own August usage at default effort. Whether your bill agrees depends entirely on how your prompts are built, which is what this guide is for.
It also untangles the naming, because Claude Fable 5.1 pricing and Claude Mythos pricing are the same rate card wearing two badges, and knowing which model you can actually buy matters more than either number.
What is Claude Fable 5.1, and how is Mythos 5.1 different?
Claude Fable 5.1 and Claude Mythos 5.1 are the same underlying model at the same price, separated by safeguards and access.
Fable 5.1 is generally available with added protections around cybersecurity, biology, and chemistry.
Mythos 5.1 runs without those restrictions and is limited to a small set of vetted organizations, currently US-based, in Anthropic’s trusted access programs. Both carry a 1-million-token context window and a 128,000-token output limit, unchanged from Fable 5.
For nearly every buyer, the practical answer is simple: you’re buying Fable. Mythos access targets verified cyberdefense and life-sciences work. Anthropic enrolled its first life-sciences participants in coordination with the US government and says it plans to widen access to the broader research community. Everyone else gets the identical capability with guardrails, billed at identical rates.
The model itself is built for long-horizon agentic work. Anthropic reports Fable 5.1 more than doubling its predecessor on Terminal-Bench-Science, from 24.7% to 52.6%, with the biggest gains on long-running tasks rather than quick completions. Anthropic says Fable 5.1 “avoids shortcuts that result in poorer-quality work, and it’s smart enough to fix the root causes of software issues.”
How Mythos got here
The Mythos tier arrived in three steps.
Claude Mythos Preview shipped in April 2026 at $25 per million input tokens and $125 per million output, restricted to approved partners in Anthropic’s Project Glasswing cybersecurity program. Mozilla used it to patch 271 Firefox vulnerabilities in two weeks.
On June 9, 2026, Anthropic released Claude Fable 5 and Claude Mythos 5 at $10/$50, a 60% cut from the Preview rate, splitting the same model into a public version with safeguards and a restricted one without.
The September 1, 2026 release of Fable 5.1 and Mythos 5.1 keeps that structure and that headline price, and moves the discount into the cache line instead.
Report
Finance needs to prove AI’s return: CloudZero report
260 senior finance leaders (more than half CFOs) told us why the speed of seeing AI spend, not the size of it, separates who pulls ahead on AI from who gets burned.
How much does the Claude Fable 5.1 API cost?
The Claude Fable 5.1 API costs $10 per million input tokens and $50 per million output tokens, with cache reads at $0.25, per Anthropic’s published pricing. The API model string is claude-fable-5-1, and the model is live on the Claude API, AWS, Google Cloud, and Microsoft Foundry.
| Fable 5.1 / Mythos 5.1 (per 1M tokens) | Rate |
|---|---|
| Input | $10.00 |
| Output | $50.00 |
| Cache reads | $0.25 |
| Cache writes (5-minute) | $12.50 |
| Cache writes (1-hour) | $20.00 |
| Batch input / output | $5.00 / $25.00 |
Two modifiers sit beside the table. US-only inference bills at 1.1x across every token category, cache writes and reads included, a 10% premium for data-locality requirements. Web search costs $10 per 1,000 searches on top of tokens, while web fetch carries no separate fee.
Cloud marketplace rates can differ from Anthropic direct, so check the broader Bedrock pricing picture if that is your route. For the rest of the model family, our Claude pricing guide has the full lineup.
Why is the Fable 5.1 cache read cut the real price cut?
Fable 5.1’s cache reads cost 0.025x the base input rate. Every other Claude model charges 0.1x, and so does GPT-6 Astra, whose cached input runs $1 against the same $10 base. In dollars: reusing a million cached tokens on Fable 5.1 costs $0.25 against $10 fresh, a 97.5% discount on repeated context.
The profile this creates is lopsided on purpose. Pay full freight and Fable runs double Opus 5. Hit the cache and you pay half of Opus’s $0.50 reads. Anthropic is effectively pricing for one workload shape: agents with large, stable context that gets reread on every turn.
The break-even math is friendly. A five-minute cache write costs $12.50 per million tokens, so one write plus one read runs $12.75 against $20 for sending the same tokens fresh twice. Caching pays from the first reuse, and every read after that is nearly free.
The catch is cache discipline. Those savings only land if prompts keep system instructions, tool definitions, and reference context in a stable prefix. Structures that shuffle content between turns break the cache and pay full freight, which is why Anthropic’s 25% to 45% figures come with an implicit asterisk labeled “if you build for it.”
Does Fable 5.1 actually cost less to run than Fable 5?
On the rate card, yes. Per completed task, the independent numbers say no. Artificial Analysis measured Fable 5.1 at $3.76 per Intelligence Index task at maximum effort, 20% more than Fable 5’s $3.14, because it generates roughly 1.7 times the output tokens. The cache cut saves about $1.40 per task, and without it Fable 5.1 would run about $5.16, yet it still loses to the verbosity. The routing comparison is sharper still: Opus 5 at maximum effort scores just three points lower at $2.34 per task.
So the two headline changes pull against each other, and which one wins depends on your workload shape. Context-heavy agents with disciplined caching land near Anthropic’s 25% to 45% savings estimates. Output-dominated work pays the verbosity tax at $50 per million, where no cache discount exists. Anthropic’s numbers and Artificial Analysis’s numbers are both true, on different workloads.
Effort levels are the lever that reconciles them. Fable 5.1’s five settings span an 11x range in output tokens, from 13.1 million at low effort to 143.7 million at maximum across the same evaluation, with scores moving from 58 to 66. Artificial Analysis found xhigh effort scoring 65 at $2.72 per task, a dollar less than maximum. Defaulting to max effort is a choice, and it’s the expensive one.
The only way to know which side of the ledger you’re on is measuring cost per workflow, not per account. Vendor savings estimates describe their benchmark workloads, not yours.
How does Fable 5.1 compare to Opus, Sonnet, and GPT-6 Astra?
Fable 5.1 costs twice Opus 5 and five times Sonnet 5 on input, and matches OpenAI’s GPT-6 Astra to the dollar. That parity, two days apart, quietly ended price shopping between frontier labs: the top of both menus now reads $10 and $50.
| Model (per 1M tokens) | Input | Output |
|---|---|---|
| Claude Fable 5.1 / Mythos 5.1 | $10.00 | $50.00 |
| GPT-6 Astra (OpenAI) | $10.00 | $50.00 |
| Claude Opus 5 | $5.00 | $25.00 |
| Claude Sonnet 5 | $2.00 | $10.00 |
A quiet gift hides in that table: Sonnet 5’s $2 and $10 introductory pricing was scheduled to rise to $3 and $15 on September 1, 2026, and Anthropic cancelled the increase. Sonnet stays the workhorse rate, which strengthens the standard routing play: send routine traffic to Sonnet or Opus and reserve Fable for the long agentic tasks that justify it.
On the OpenAI side, the GPT-5.6 lineup maps OpenAI’s cheaper tiers. If you run Fable through Claude Code, that pricing works differently and has its own guide.
What are the pros and cons of Fable 5.1’s pricing?
Where the rate card works for you:
- Cache reads at $0.25 per million are the cheapest on any Claude or OpenAI flagship, at 0.025x against the 0.1x both lineups otherwise charge
- Caching breaks even on the first reuse: $12.75 for a write plus a read against $20 fresh
- Five effort levels span an 11x output-token range, turning one model into a price ladder
- Batch halves input and output to $5 and $25 for asynchronous work
- No price increase over Fable 5, while its Terminal-Bench-Science score more than doubled in Anthropic’s testing
Where it works against you:
- Output still bills $50 per million with no cache relief, and Artificial Analysis measured 20% higher cost per task at maximum effort despite the cache cut
- Fresh input costs double Opus 5 and five times Sonnet 5, so misrouted routine traffic is expensive
- The advertised 25% to 45% savings require cache-preserving prompt architecture most teams haven’t built
- US-only inference adds 1.1x across every token category, and web search adds $10 per thousand calls inside agent loops
- Mythos 5.1 access is closed to all but vetted organizations, so the unrestricted variant isn’t a purchasable option
How do you keep Fable 5.1 spend under control?
Treat the cache as infrastructure, route by effort level, and measure per workflow. Concretely: stabilize prompt prefixes so cache reads do the work Anthropic priced them for, move asynchronous jobs to Batch, set effort per use case instead of defaulting to maximum, and keep routine traffic on Sonnet where the cancelled price increase just made it a better deal.
Then verify the savings actually arrived, because this rate card makes account-level billing useless. Cached and fresh tokens, five effort levels, batch and standard tiers, and a 1.1x residency modifier all collapse into one invoice line unless allocation happens at the level where decisions get made.
That allocation is what CloudZero does with Anthropic usage: our Anthropic integration puts Claude spend next to the rest of your cloud and AI spend, and the CloudZero and Anthropic partnership goes deeper than billing exports. Cache hit rates are an engineering metric until they become a unit cost. Then they’re a budget conversation you can actually win.
If Fable 5.1 is about to become a real line item, book a demo and we’ll model it against your actual workloads.
Claude Fable 5.1 and Mythos pricing FAQs
