Tom Hundley
On June 9, 2026, Anthropic resolved the cliffhanger that has hung over its Project Glasswing program since April: the Mythos-class capability it built but refused to release is now generally available — with guardrails. Today the company shipped two models on the same underlying system. Claude Fable 5 is the public one, "a Mythos-class model that we've made safe for general use." Claude Mythos 5 is the vetted-only one — the same model "with the safeguards lifted in some areas," restricted to Project Glasswing partners and a forthcoming trusted-access program.
That distinction — Fable broadly available, Mythos still locked down — is the entire story. For two months the open question in this series was whether Anthropic would ever let the public touch a model that can autonomously find and exploit quarter-century-old vulnerabilities. The answer, as of today, is: yes, but only the version that won't help you do the dangerous part. The model is the same. The safeguards are the only difference.
A framing note, as in every piece in this series: Elegant Software Solutions did not participate in Project Glasswing and has no inside access to Mythos. Everything below is drawn from Anthropic's own announcement page and corroborating press coverage.
Mythos's headline capability was always going to force a release decision. Back in April, Anthropic published a full system card and a price tag for Claude Mythos Preview — then declined to sell it, calling the non-release a voluntary dual-use choice rather than a tripped safety threshold. The catch was explicit: general access was gated on safeguards that did not yet exist. Today's launch is Anthropic declaring those safeguards built and, in its words, "now robust enough for a general release."
The significance is not a new model in the benchmark sense — Fable 5 and Mythos 5 share one underlying system. It is that a deliberately withheld frontier capability has, for the first time, partly entered the open market. Anthropic puts the bar plainly: "Fable 5's capabilities exceed those of any model we've ever made generally available." This is the first Mythos-class model the public can actually use.
The name carries the lineage. Per Anthropic: "Fable is from the Latin fabula, 'that which is told,' akin to the Greek mythos." Same root, same family — a public sibling to the model that started the story.
It is worth being exact, because the two-model framing invites confusion. Fable 5 and Mythos 5 are not a strong model and a weak model. They are one model deployed under two safety postures.
Claude Fable 5 is the full-capability model with a safety layer bolted on. In three high-risk domains, its classifiers intercept the request and hand it to a less capable model — "Requests that are flagged by our classifiers as being part of such distillation attempts will fall back to Opus 4.8." The same fallback applies across all three safeguarded areas: when a guardrail fires, you get Claude Opus 4.8's answer, not Fable 5's.
Claude Mythos 5 is "the same underlying model as Fable 5, but with the safeguards lifted in some areas." It is not for sale to the public. It will "initially be deployed through Project Glasswing, in collaboration with the US government, as an upgrade to Claude Mythos Preview," and Anthropic states flatly that "it has the strongest cybersecurity capabilities of any model in the world." Access splits two ways:
A broader trusted-access program is planned. But the headline for general readers is simple: you can use Fable 5; you cannot use Mythos 5.
Anthropic's safeguards target three domains where the underlying model's capability is genuinely dual-use:
How often do these fire? Anthropic says the safeguards are deliberately conservative — "they'll sometimes catch harmless requests" — and "they trigger, on average, in less than 5% of sessions." For the overwhelming majority of users, then, Fable 5 simply behaves like a top-tier model. TechCrunch made the same point from the skeptic's side, framing the launch as Anthropic releasing "its most powerful model publicly days after warning AI is getting too dangerous" — with the guardrails as the only thing between the public and the unrestricted version.
Fable 5 is available now, with a rollout that differs by plan:
The pricing headline is a genuine surprise that cuts against the usual frontier trajectory. Both Fable 5 and Mythos 5 cost $10 per million input tokens and $50 per million output tokens — which Anthropic notes is "less than half the price of Claude Mythos Preview." At its April launch, that preview was quoted at $25 per million input and $125 per million output. A more capable, generally available model arriving at well under half the withheld preview's price is the opposite of the premium-for-frontier pattern, and the most consequential commercial fact in today's release for anyone budgeting around these models.
Anthropic claims state-of-the-art performance across nearly all the benchmarks it tested, but the customer evidence is the more credible spine — these are named partners describing production workloads, not internal scores.
Stripe reported the standout result: "Fable 5 compressed months of engineering into days. In a 50-million-line Ruby codebase, the model performed a codebase-wide migration in a day that would otherwise have taken a whole team over two months by hand." On Cognition's FrontierCode, Fable 5 posted the highest score among frontier models at medium effort. Three named platform leaders backed the long-horizon claim on the record:
"Claude Fable 5 is the state of the art model on CursorBench. It's opened up a class of long-horizon problems that were out of reach for earlier models."
— Michael Truell, CEO and Co-founder, Cursor
"Claude Fable 5 is a real step forward for the developers GitHub serves. In our early testing, it took on complex, long-horizon coding tasks with a level of autonomy and reliability that exceeded previous benchmarks."
— Mario Rodriguez, Chief Product Officer, GitHub
"Claude Fable 5's reasoning is a clear step beyond Opus 4.8. It works at senior research scientist grade — picking directions, allocating resources, killing its incorrect beliefs, and producing novel first-principles outputs."
— Sean Ward, CEO and Co-founder, Applied Intuition
On vision, Anthropic reports that earlier Claude models needed a complex helper harness to play Pokémon, while Fable 5 completed FireRed with a minimal, vision-only setup — a real step up in pure visual reasoning. On a persistent-memory task built around Slay the Spire, the model's performance improved "three times more" and it reached the final act "three times more often" once it could carry memory across a run. And on knowledge work, Hebbia reported Fable 5 posted "the highest score of any model" on its Finance Benchmark, with the company's AI Research Lead, Izzy Miller, adding that it was the "first to break 90% on our core analytics benchmark...a 10-point jump over Opus."
The bio results belong to Mythos 5, the version with biology and chemistry safeguards removed, and they are the clearest illustration of why that version stays vetted-only. Anthropic's internal experts reported that Mythos 5 "accelerated aspects of the drug design process by around ten times," and that "nine of the 14 protein targets...yielded strong candidates for drug design." In blind comparison, scientists "preferred Mythos's molecular biology hypotheses ~80% of the time." That is exactly the capability Anthropic is gating: enormously valuable for legitimate drug design, and dual-use enough that the public model routes the sensitive steps to Opus 4.8 instead.
This release is best read as the next chapter of Project Glasswing, not a standalone launch. Glasswing is Anthropic's coalition "in collaboration with the US government" to put Mythos-class models in the hands of cyberdefenders and critical-software-infrastructure providers. The sequence is now clean:
Anthropic also notes Glasswing's continued expansion to "approximately 150 new organizations in more than fifteen countries." Read against the earlier chapters of this series, today's launch resolves the explicit gap the preview left open: the system card said general release waited on safeguards Anthropic had not built. As of June 9, the company says it has built them — and the proof is that the public now gets the model, just not the parts that made withholding it the responsible call.
That is the honest summary of where Glasswing stands. The withheld capability is no longer fully withheld. It is split in two: a public model that refuses the dangerous 5%, and a vetted one that doesn't refuse at all.
Can I use Fable 5 today?
Yes. Claude Fable 5 is available now on the Claude API and consumption-based Enterprise plans. It is also included at no extra cost on Pro, Max, Team, and seat-based Enterprise plans from June 9 through June 22; starting June 23 it requires usage credits on those plans, with standard-plan restoration when capacity allows.
What's the difference between Fable 5 and Mythos 5?
They are the same underlying model. Fable 5 has safeguards that block three high-risk domains and fall back to Claude Opus 4.8; Mythos 5 has those safeguards lifted in some areas. Fable 5 is broadly available. Mythos 5 is restricted to Project Glasswing partners (cyber safeguards lifted) and selected biomedical researchers via a trusted-access program (biology/chemistry lifted, cyber safeguards intact).
How much does it cost?
Both models are priced at $10 per million input tokens and $50 per million output tokens — which Anthropic notes is less than half the price of the April Mythos Preview (which launched at $25 input / $125 output per million tokens).
What won't Fable 5 do?
It blocks and routes to Opus 4.8 in three areas: cybersecurity (exploitation and offensive/agentic tasks), biology and chemistry (dual-use work such as AAV design), and distillation (extracting its capabilities to train competing models). Anthropic says these conservative safeguards trigger in less than 5% of sessions on average and sometimes catch harmless requests.
Why release the public version now, after withholding Mythos?
Anthropic withheld Mythos Preview in April pending safeguards it had not yet built. It now says those safeguards are "robust enough for a general release," so the safe, safeguarded version (Fable 5) can go public while the unguarded version (Mythos 5) remains vetted-only.
Is Fable 5 actually more capable than Anthropic's previous public models?
Anthropic claims state-of-the-art performance on nearly all benchmarks it tested and says Fable 5's capabilities "exceed those of any model we've ever made generally available." Named customers corroborated the long-horizon coding claims (Stripe, Cursor, GitHub, Applied Intuition, Cognition, Hebbia); these are vendor- and Anthropic-reported figures, not independent evaluations.
Discover more content:
