On June 9, 2026, Anthropic released Claude Fable 5 — making a model of the so-called Mythos class generally available for the first time. In Anthropic's hierarchy, the Mythos class sits above the Opus line; until now it was reserved for a small group of cyber defenders and infrastructure providers through Project Glasswing. Fable 5 changes that.
The name says it all: Fable comes from the Latin fabula — "that which is told" — akin to the Greek mythos. Same model, two names: Claude Mythos 5 (without cyber safeguards, vetted partners only) and Claude Fable 5 (with safeguards, for everyone). This post takes a sober look at what the model can do, what it costs and what to watch out for.
What is Claude Fable 5?
According to Anthropic, Fable 5 (claude-fable-5) is state of the art on nearly all tested benchmarks — with particular strengths in software engineering, knowledge work, vision and scientific research. The core claim: the longer and more complex the task, the bigger the lead over previous models. On some benchmarks Fable 5 scores more than 10% above Claude Opus 4.8, which itself only shipped in late May.
A few early-access results that stand out:
- Stripe reports that Fable 5 completed a migration in a 50-million-line codebase in a single day — work that would otherwise have taken a whole team over two months.
- On Cognition's FrontierCode benchmark (production-grade coding tasks), Fable 5 scores highest among frontier models — even at medium reasoning effort.
- In vision, the model is the new state of the art: it beat Pokémon FireRed end to end using only raw screenshots, with none of the helper harnesses earlier models needed.
- On Hebbia's finance benchmark for senior-level reasoning, it posts the highest score of any tested model.
The technical specs
| Feature | Claude Fable 5 |
|---|---|
| Class | Mythos (above Opus) |
| API ID | claude-fable-5 |
| Context window | 1,000,000 tokens |
| Modalities | text + image + files → text |
| Reasoning | yes, with controllable effort |
| Price (API) | $10/M input · $50/M output |
| Safeguard fallback | Claude Opus 4.8 (< 5% of sessions) |
| Release | June 9, 2026 |
Token efficiency is notable too: several early-access customers report that Fable 5 solves tasks in fewer turns and with fewer reasoning tokens than its predecessors — so the higher per-token price is partly offset by lower consumption per task.
The safeguards: why there are two models
A model this capable carries misuse risks — especially in cybersecurity. Anthropic's answer: separate safety classifiers that detect requests related to cybersecurity, biology/chemistry and model distillation. When such a request is detected, Claude Opus 4.8 answers instead of Fable 5 — and the user is informed. That's far better than a hard refusal, and Opus 4.8 is a top-tier model in its own right.
Anthropic deliberately tuned the classifiers conservatively: occasionally harmless requests get rerouted too. But according to Anthropic, more than 95% of all sessions run with no fallback at all — and there, Fable 5 is effectively identical to the unrestricted Mythos 5. An external bug bounty found no universal jailbreak in over 1,000 hours of testing.
Privacy: the 30-day rule
Important for use in Europe: Anthropic requires a mandatory 30-day retention of requests for all business traffic on Mythos-class models — exclusively for abuse detection. The data is not used for training, every human access is logged, and it is deleted after 30 days. Anyone deploying Fable 5 in production should disclose this transparently in their privacy policy — we've already done so for Zeptix in the privacy policy and the subprocessor list.
Pricing: premium, unapologetically
At $10/M input and $50/M output tokens, Fable 5 costs twice as much as Opus 4.8 — clearly positioned as a premium model. Anthropic's own framing is interesting: less than half the price of the Mythos Preview. Early customers still report lower cost per task, because the model needs fewer attempts. So the real question isn't "more expensive per token" but "what does the finished result cost".
Using Claude Fable 5 in Zeptix
Fable 5 is available in every Zeptix dashboard's model picker as of today — exclusive to the Business plan. A response costs 80 credits per message in language bots; coding bots bill by length (roughly 267 credits per request on average — for comparison: GPT-5.5 sits at 157, Opus 4.8 at 133). The model picker points out the 30-day retention and the safeguard fallback directly.
Activation steps and all details are in the announcement: Claude Fable 5 is live — Anthropic's Mythos class for Business bots.
Who should switch? If your bot handles complex, multi-step tasks — demanding expert advice, long document analysis, agentic coding across large codebases — Fable 5 is the new reference. For classic FAQ and support bots, cheaper models like Claude Haiku 4.5 or Gemini 2.5 Flash usually remain the more economical choice. As always: run a quick A/B test with your typical requests, then decide.
Conclusion
Claude Fable 5 is more than a version bump: it's the first time a model above the Opus class is usable by everyone — with thoughtful safeguards instead of blanket bans. The price is confident, and judging by the first tests, so is the performance. With Fable 5 selectable in Zeptix starting today, Business customers get a new lever for their most demanding use cases.
Curious? Build a chatbot for free and see what the Mythos class does for your use case.
Build your own chatbot
Upload your knowledge, pick your model and go live in minutes — no code.
