Claude Mythos 5 and Fable 5 Explained
Summary. Mythos 5 and Fable 5 are two versions of the same underlying Claude model, released by Anthropic on June 9, 2026. They sit in a new top tier called Mythos-class, above the Opus tier. The only real difference between them is the safety controls applied on top. In June 2026 both models were briefly suspended because of US export controls, not because the models themselves were "banned" as products. Access was restored on July 1, 2026.
Note: This document describes a fast-moving topic. Details may have changed after the date above. For the latest official information, check Anthropic's website directly.
1. What Is a "Mythos-Class" Model?
Anthropic organises its Claude models into tiers by capability:
| Tier | Purpose |
|---|---|
| Haiku | Fastest and cheapest; good for high-volume, simple tasks. |
| Sonnet | Balanced tier; a common choice for production apps. |
| Opus | High capability for hard, complex tasks. |
| Mythos | The new top tier, above Opus. Built for long, multi-step agentic work. |
Mythos-class models are designed for agentic use: long reasoning chains, reliable tool use, and staying coherent across very long tasks. The tier first appeared in April 2026 with Claude Mythos Preview, released to a small group of partners through a program called Project Glasswing.
2. Mythos 5 vs Fable 5 — What Is the Difference?
This is the key point that confuses many people. They are the same model. The weights, the training, the capabilities, and the price are identical. Only the safeguards differ.
| Aspect | Claude Fable 5 | Claude Mythos 5 |
|---|---|---|
| Underlying model | Same shared model | Same shared model |
| Safeguards | Full safety classifiers active | Some safeguards lifted (e.g. cybersecurity) |
| Who can use it | General public (API and Claude apps) | Only trusted Project Glasswing partners |
| Main use case | Safe general-purpose use | Defensive cybersecurity work |
| Pricing | $10 / 1M input, $50 / 1M output | Same |
A good way to think about it: it is one engine, offered in two ways — one with a governor installed (Fable 5), and one with that governor removed for a small set of vetted users (Mythos 5). The names come from the same root idea: fable is from the Latin fabula ("that which is told"), close in meaning to the Greek mythos.
How Fable 5's safeguards actually work
Fable 5 does not usually refuse a sensitive request. Instead, it reroutes it. Separate classifier models watch each incoming request. If a request touches a covered area, the answer is produced by Claude Opus 4.8 instead, and the user is told this happened.
The covered areas are cybersecurity, biology and chemistry, and distillation (attempts to copy Claude's capabilities to train competing models). According to Anthropic, more than 95% of Fable 5 sessions run entirely on Fable 5 with no fallback at all.
3. Why Were the Models Suspended? (The "Ban" Story)
The word "banned" is used loosely in the news, so it helps to be precise. The models were not banned as unsafe products. They were suspended because of a US government export-control order. Here is the timeline.
What triggered it
Amazon researchers reported a jailbreak — a way to bypass one of Fable 5's cybersecurity safeguards. In one case, the model produced code showing how a software vulnerability could be exploited. This report was flagged to the US government, which then applied export controls: rules that limit which foreign nationals may access a technology.
Why access went dark for everyone
The order took effect immediately and told Anthropic to block access by any foreign national, inside or outside the US. Because Anthropic had no reliable way to verify nationality in real time, it chose to suspend access for all users rather than risk breaking the rule.
How it was resolved
Anthropic worked with the US government and Amazon to review the report. Their testing found that many less capable models — including Opus 4.8 and several non-Anthropic models — could identify the same vulnerabilities, so the technique did not reveal a unique Mythos-level capability. It was described as a borderline case for the safeguards. Anthropic then trained an improved safety classifier that blocks the reported behaviour in more than 99% of cases. On June 30 the Department of Commerce lifted the controls, and Fable 5 returned globally on July 1.
Current status: Fable 5 is available to the public again. Mythos 5 remains limited to a small number of approved US organizations (around 100 vetted groups focused on critical infrastructure and cybersecurity) through Project Glasswing.
4. Key Facts at a Glance
| Question | Answer |
|---|---|
| What tier are they? | Mythos-class — Anthropic's top tier, above Opus. |
| Are they the same model? | Yes. Only the safeguards differ. |
| Which one can I use? | Fable 5 (public). Mythos 5 is limited to vetted partners. |
| Why were they suspended? | A US export-control order, not a product safety recall. |
| Are they available now? | Fable 5: yes, globally. Mythos 5: limited access only. |
5. Why This Matters for Engineers
If you build on the Claude API, the practical points are simple:
- Handle refusals and fallback. On Fable 5, a blocked request can return a refusal or be routed to Opus 4.8. Plan your integration to detect this and retry on another model if needed.
- Treat frontier models as infrastructure. The suspension showed that access can be affected by policy decisions, not only technical issues. It is wise to have a fallback model ready in case any single model becomes unavailable.
- Check the official docs for current status. Pricing, availability, and safeguard behaviour can change quickly.