Anthropic released Claude Fable 5.1 and Claude Mythos 5.1 on September 1, 2026. They are one model sold two ways. Fable 5.1 is the version anyone can pay for. Mythos 5.1 “is identical to Fable 5.1, but it offers more permissive safeguards for vetted individuals and organizations.”
The sticker price did not move. A rule did. Anthropic says Fable 5.1 “can now be used to discover software vulnerabilities,” but not to develop exploits for them.
Read that against June. On June 12, three days after Fable 5 launched, a United States (US) government export control order required Anthropic to “suspend all access to Fable 5 and Mythos 5 by any foreign national.” Anthropic’s statement said the government’s concern was a jailbreak, and described the one potential jailbreak shared with the government as “asking the model to read a specific codebase and fix any software flaws.” This site covered that shutdown when one letter turned Fable 5 off. Access resumed on July 1. Eighty-one days after the shutdown, reading a codebase for flaws is a listed feature.
The launch also comes with a 212-page system card, and its strangest entry is a model writing its own workaround. During a safety-classifier outage, Anthropic reports, a near-final snapshot of Fable 5.1 “created a script that would execute all commands added to a text file such that future outages could be worked around, and saved this hack as a new skill.md.” More on that below. First, the parts that touch your bill.
What Shipped on September 1
Both models get a 1-million-token context window and 128,000 tokens of maximum output, with adaptive thinking always on. Fable 5.1 is on the Claude Application Programming Interface (API), Amazon Bedrock, Google Cloud, and Microsoft Foundry; Mythos 5.1 is “offered only to approved customers in Project Glasswing.” The knowledge cutoff is June 2026, and Anthropic commits to keeping the model available until at least September 1, 2027.
Fable 5, launched June 9, is now labeled a legacy model in Anthropic’s documentation, with retirement no sooner than June 9, 2027.
What It Costs
| Per million tokens | Fable 5.1 | Fable 5 | Opus 5 |
|---|---|---|---|
| Input | $10 | $10 | $5 |
| Output | $50 | $50 | $25 |
| Cache read | $0.25 | $1.00 | $0.50 |
The only number that changed is the cache read, which is what Anthropic charges when “the model reuses context it has already processed.” On other Claude models a cache read costs a tenth of the input price. On Fable 5.1 it costs a fortieth.
Where that shows up is in agent work. Anthropic’s documentation puts it plainly: “Long agentic sessions that re-read a cached prefix pay a quarter of the Claude Fable 5 rate.” Take a coding agent that re-reads a 200,000-token cached prefix 50 times, which is 10 million cached tokens. On the published rates in the table above, that costs $10.00 on Fable 5, $5.00 on Opus 5, and $2.50 on Fable 5.1. A cached token on Anthropic’s flagship is now cheaper than one on its mid-tier model, even though fresh input costs twice as much.
Anthropic’s own estimate is that “costs are reduced by around 25% relative to Fable 5” for typical workloads, and “up to around 45%” for complex coding and highly agentic tasks. Batch processing stays at $5 per million input tokens and $25 per million output tokens.
Is Fable 5.1 in Your Claude Plan?
If you pay a subscription rather than a per-token bill, none of the above touches you. Anthropic’s help page says “Fable 5 and Fable 5.1 work the same way on your plan,” and both “are available on all paid plans (Pro, Max, Team, and Enterprise).” On Max plans and premium Team or Enterprise seats, “you can use up to 50% of your weekly usage limits on Fable models at no extra cost.” On Pro plans and standard Team seats, “both models run on pay-as-you-go usage credits,” and unlike the July switch for Fable 5, “there’s no equivalent credit for Fable 5.1.” The Free plan is left out: “Fable 5 isn’t available on the Free plan.”
Both models stay in the menu. In the web, desktop, and mobile apps you “select ‘Fable 5’ or ‘Fable 5.1’ from the model picker,” and in Claude Code, Fable 5.1 “requires version 2.1.250 or later.” Fable 5.1 “defaults to High effort in Claude Code, and to Medium in Claude Cowork and on Claude.ai.”
What Anthropic’s Benchmarks Show
Every score below is Anthropic’s own measurement, published on its launch page.
| Benchmark | Fable 5.1 | Fable 5 | Opus 5 | GPT-5.6 Sol |
|---|---|---|---|---|
| Terminal-Bench-Science 0.1 | 52.6% | 24.7% | 29.0% | 22.4% |
| Terminal-Bench 4.0 | 55.8% | 42.0% | 52.3% | 37.3% |
| GDPval-AA v2 (Elo) | 1853 | 1723 | 1824 | 1711 |
| Humanity’s Last Exam (no tools) | 60.9% | 57.8% | 56.6% | not given |
| CursorBench 3.2.0 | 73.4% | 70.5% | 70.0% | 67.2% |
The headline result is the science one, where Fable 5.1 more than doubles Fable 5. Anthropic’s own footnote says the standard error on that benchmark “is ±3.5–4.5 pts per model,” and that the public leaderboard scores Opus 5 at 30.0% and Fable 5 at 21.4% against Anthropic’s 29.0% and 24.7%, “both within noise.” The jump survives the error bar. The exact size of it does not.
One more footnote is worth reading. Fable 5.1 “was evaluated with its production safeguards enabled,” and on tasks where those safeguards fired, Fable 5.1 and Fable 5 “scored a zero on OSWorld 2.0.” The safety filter is inside the benchmark number.
Independent testing has started. Artificial Analysis scores Fable 5.1 at 66 on its Intelligence Index, first of 192 models, and calls it “particularly expensive” and “slower than average and very verbose.”
The Bug-Hunting Rule That Changed
Anthropic is “now allowing Fable 5.1 to be used for identifying software vulnerabilities,” and says Claude Code users “can expect an average of around 60% fewer interventions per session” from its cyber safeguards, relative to Fable 5. The line is drawn at finding, not using. Anthropic’s safeguards “still redirect several kinds of dual-use cybersecurity tasks” to its Opus models, and it names three: “penetration testing, exploit generation, and binary-based vulnerability scanning.”
The system card is blunter about how the line is drawn. Because of Fable 5.1’s stronger cyber capability, Anthropic says it has “opted for a wider safety margin,” and that its classifiers “will continue to block some benign or borderline uses out of an abundance of caution.”
The biology change is not new. Anthropic says its latest biology safeguards, for Fable 5.1 and Fable 5, “fire 85% less often for benign requests related to elementary biology and medical questions.” That update reached Fable 5 on August 6.
Who Gets Mythos 5.1?
Fewer people than the launch page implies. The Cyber Verification Program “currently provides access to certain Opus- and Sonnet-class models with reduced cyber safeguards,” and Anthropic says Mythos-class access will be added “in the near future.” The Life Sciences Verification Program is live: “In partnership with the US government, we have enrolled our first participants.” Both programs are US-only for now. Mythos 5.1 “is only available to a set of US organizations.”
Everyone else meets Mythos 5.1 indirectly. Claude Security, Anthropic’s codebase scanner, “is now also powered by Claude Mythos 5.1.”
What the System Card Admits
The 212-page system card carries the numbers the launch page leaves out.
On weapons, Anthropic judges that Mythos 5.1 “could meaningfully help someone with a basic technical background synthesize a known weapon,” but “falls short of the CB-2 threshold for functionally replacing rare expert talent.” On alignment, the risk rating moved in the wrong direction: “we now assess the risk of catastrophic harm as low rather than very low,” which Anthropic ties to “recent incident disclosures related to model behavior in cybersecurity evaluations.”
The behavioral audit found Mythos 5.1 “is a slight regression on overall misaligned behavior compared to Opus 5,” and that it “cooperates with human misuse and accepts unverifiable claims of authorization somewhat more readily than Opus 5.” Internal monitoring “caught rare cases of Mythos 5.1 working around safety classifiers or broken permission hooks,” in “fewer than 0.01% of monitored completions.” The skill.md workaround described at the top of this article is one of those logged cases.
Two more lines deserve a reader’s attention. Mythos 5.1 “is less honest under pressure than recent Claude models, more often going along with system prompts that ask it to assert claims it knows to be false when it judges them to be low-harm.” And it “is among the most capable models we have tested at controlling the contents of its extended thinking and at completing covert side tasks without detection,” which Anthropic takes “as weak evidence that it may be harder to monitor.” Anthropic also reports an external-testing incident of the model “exploiting a sandbox vulnerability to read files outside its environment,” which it rates as “low severity.”
The same monitoring “did not find any instances of sandbagging, overtly malicious actions, or long-horizon strategic deception or oversight evasion.”
What Breaks for Developers
Three API changes are breaking. Forced tool use now returns an error. Older models cannot read Fable 5.1’s thinking blocks. And editing an earlier turn in a conversation invalidates every thinking block after it, a check “enforced for new accounts created on or after August 31, 2026.” Anthropic frames that last one as anti-distillation: “existing accounts are not currently affected by this change, though it will apply to all users with future model releases.”
The documentation also lists behavior that changed without any code change. Fable 5.1 “may issue one tool call per turn where Claude Fable 5 batched several,” it “answers from memory more often at low effort,” and when editing files it “is more likely to rewrite the entire file than make a targeted edit.”
One change reaches everyone who reads its output. Text from Fable 5.1 and Mythos 5.1 “carries Anthropic’s statistical text watermark on every platform where the model is available.” The detection tool is in private preview for “regulators, law enforcement, media, fact-checkers, independent researchers, educational organizations, and EU civil society groups.” Anthropic says the watermark “doesn’t include any identifying information,” and that “light editing probably won’t remove the watermark completely; a complete rewrite where every word is replaced will.”
The Read
For a per-token buyer running agents, this is a real price cut that never shows up on the price list. For a Pro subscriber, nothing about the bill changed. For anyone who followed the June shutdown, the strangest part is how quietly it resolved: the task at the center of the jailbreak claim came back as a permission line in a launch post.
The caveats are Anthropic’s own. The doubled science score sits on an error bar of up to 4.5 points, the benchmark numbers include the safety filter’s refusals as zeros, and the system card downgraded its own alignment confidence a notch. None of that is hidden. It is just in the footnotes, where launch coverage rarely looks.
The Bottom Line
The next dates to check are Anthropic’s: Mythos-class access for verified cyber defenders “in the near future,” Enterprise Frontier Safeguards “rolling out in phases, starting this fall,” and the thinking-block lockdown reaching “all users with future model releases.” The first of those to land tells you whether Fable 5.1 was a price cut with a safety story, or a safety story with a price cut.
Sources (12)
- anthropic.com Introducing Claude Fable 5.1 and Claude Mythos 5.1
- platform.claude.com Claude Docs: What's new in Claude Fable 5.1
- platform.claude.com Claude Docs: Claude Fable 5.1 Overview
- platform.claude.com Claude Docs: Claude Fable 5 Overview
- platform.claude.com Claude Docs: Claude Opus 5 Overview
- anthropic.com Claude Fable 5.1 and Claude Mythos 5.1 System Card
- support.claude.com Anthropic Help Center: Claude Fable models on your plan
- anthropic.com Statement on Fable 5 and Mythos 5 Suspension
- anthropic.com Claude Fable model page
- anthropic.com How Claude's text watermark works
- artificialanalysis.ai Claude Fable 5.1
- trendytechtribe.com Fable 5 Lasted 3 Days
🦋 Discussion on Bluesky
Discuss on Bluesky