Anthropic shipped Claude Fable 5 on June 9—the first Mythos-class model made safe for general use. It's the capability tier that sat behind Project Glasswing since April, now opened to everyone with a new safeguard architecture: instead of refusing risky queries outright, Fable 5 routes them to Claude Opus 4.8. The longer and more complex the task, the larger Fable's lead over every other generally available Claude model—and early customer tests put it at the top of coding, vision, and long-horizon agent benchmarks.
What Mythos-Class Means in Practice
Mythos-class models sit above Opus in Anthropic's lineup. Fable 5's capabilities exceed anything Anthropic has ever released publicly—state-of-the-art on nearly every benchmark the company tested, with the gap widening as tasks get longer and harder. The headline engineering example: Stripe reported that Fable 5 compressed months of work into days, completing a codebase-wide migration across a 50-million-line Ruby monolith in a single day—work that would have taken a team more than two months by hand.
On Cognition's FrontierCode evaluation—which tests whether models can pass difficult coding tasks while meeting production-codebase standards—Fable 5 scores highest among frontier models, even at medium effort. Cursor calls it state of the art on CursorBench and says it opens a class of long-horizon problems that were out of reach for earlier models. Cognition reports the highest score on FrontierBench. On ViBench, an end-to-end vibe-coding benchmark, Fable 5 nearly saturates base use cases and builds apps in less time with fewer tokens.
Safety Routing, Not Refusal
Releasing a model this capable without safeguards would be reckless. Fable 5's answer is a set of safety classifiers—separate AI systems that detect potential misuse and jailbreak attempts. When a query triggers a classifier, the response is handled by Opus 4.8 instead of Fable 5, and users are told when the fallback happens.
Three areas are covered:
- Cybersecurity — exploitation, offensive cyber tasks, and agentic hacking workflows.
- Biology and chemistry — broadly tuned for now; Anthropic says it will narrow these safeguards as it refines them.
- Distillation — attempts to extract Fable's capabilities to train competing models.
The routing design is deliberate. A fallback to Opus 4.8 is a far better experience than a hard refusal—and Opus 4.8 is itself a highly capable model. Anthropic's early data shows that more than 95% of Fable sessions involve no fallback at all; for those sessions, performance is effectively the same as the unrestricted Mythos 5 variant.
Red-teaming was extensive. An external bug bounty produced no universal jailbreaks in over 1,000 hours of testing. One external partner found Fable 5 complied with zero harmful single-turn cyber requests across 30 public jailbreak techniques. Anthropic acknowledges the safeguards are tuned conservatively and will sometimes catch harmless requests—frustrating, but intentional while they iterate post-launch.
Claude Mythos 5 for Glasswing Partners
Alongside Fable 5, Anthropic launched Claude Mythos 5—the same underlying model with cyber safeguards lifted—for Project Glasswing partners. All users with Mythos Preview access can upgrade today. Mythos 5 is comparable to or stronger than Mythos Preview in most cases, at less than half the price of the preview tier.
Anthropic plans to expand Mythos 5 through a broader trusted access program, with a separate biology track coming for researchers who need chemistry safeguards removed. For everyone else, Fable 5 is the generally available Mythos-class option via claude-fable-5 on the API and claude-mythos-5 for approved Glasswing customers only.
Long-Horizon Capabilities
Fable 5 can work autonomously longer than any previous Claude model. A few capabilities that matter for developers:
- 1M token context with improved memory—persistent file-based notes improved Slay the Spire performance three times more than Opus 4.8 did, and Fable reached the game's final act three times more often.
- Vision — state of the art; Fable beat Pokémon FireRed with a minimal vision-only harness where earlier Claude models needed complex helper tooling. It can also rebuild a web app's source code from screenshots alone.
- Token efficiency — more capable engineering in fewer turns; one customer reports spreadsheet tasks finishing 25–30% faster than Opus 4.8 at every effort level.
- Self-validation — at highest effort, Fable reflects on and validates its own work, which customers cite as what makes highly autonomous operations possible.
API support at launch includes effort control, task budgets (beta), the memory tool, context editing, compaction, and vision—same surface as Opus 4.8 plus the model's longer-running autonomy.
30-Day Data Retention
New policy for Mythos-class models: Anthropic now requires 30-day retention for all traffic on Fable 5, Mythos 5, and future models at this capability tier—on both first- and third-party surfaces. The data won't be used to train new Claude models or for non-safety purposes; human access is logged and data is deleted after 30 days in almost all cases. The retention exists to defend against complex multi-request jailbreaks and to reduce classifier false positives.
Enterprise teams evaluating Fable 5 should factor this into their compliance review alongside the capability gains.
Pricing and the June 22 Window
Both Fable 5 and Mythos 5 are priced at $10 per million input tokens and $50 per million output tokens—less than half what Mythos Preview cost. The API and consumption-based Enterprise plans have full availability from day one.
For subscription users, Anthropic is rolling out access in stages because demand is hard to predict:
- Through June 22: Fable 5 is included at no extra cost on Pro, Max, Team, and seat-based Enterprise plans.
- From June 23: Fable 5 drops from those plans; using it requires usage credits unless Anthropic extends the window.
- Eventually: Anthropic intends to restore Fable 5 as a standard subscription feature once capacity allows.
That's a two-week free trial on every paid Claude plan—enough time to run a real migration or multi-day agent workflow and decide whether the token price is worth it before credits kick in.
Why It Matters for Web Developers
The practical read: Fable 5 is the model you reach for when Opus 4.8 hits a wall—multi-day migrations, screenshot-to-code rebuilds, apps that used to take a hundred prompts and now one-shot. Point your API calls at claude-fable-5 and you get Mythos-class reasoning with Opus as the safety net, not a hard stop.
Three moves this week. First, use the free window through June 22 to stress-test Fable on your hardest backlog item—the Stripe-scale migration story isn't marketing fiction; it's the class of work this model was built for. Second, audit your compliance posture for the 30-day retention policy before routing production customer data through Fable. Third, keep Opus 4.8 in your stack—it's now the fallback layer inside Fable's safety architecture, and for most sessions you'll never notice the handoff. The frontier just moved up a tier, and for the first time, developers outside Glasswing get to use it.