Claude Fable 5 vs Mythos 5: Same Weights, Different Access
Fable 5 and Mythos 5 are the same underlying model with different safeguards. What actually differs, who can use each, and which one you are really calling.
Fable5 Editorial
People keep searching "Claude Fable vs Mythos" expecting two different models. They are not \u2014 they are the same weights with two release configurations. Here is exactly what differs and why it matters.
Update — July 1, 2026: Both models are back. Fable 5 was restored worldwide on July 1 after the US export-control directive was withdrawn on June 30 — API, Claude.ai, Claude Code, Cowork. Through July 7, Pro/Max/Team plans include Fable 5 in up to 50% of the weekly usage cap; after that it moves to usage credits ($10/$50 per million tokens). Mythos 5 was approved June 26 and resumed July 1 for approved US organizations under Project Glasswing — still no public access. Live status: /is-fable-5-back · Full story: Claude Fable 5 is back.
Fable 5 vs Mythos 5: the short answer
Same model, different safeguards and access. Anthropic split one Mythos-class release into two: Claude Fable 5 is the safeguarded version anyone can use; Claude Mythos 5 is the same underlying model with safeguards lifted in some areas, reserved for a small group of vetted cyberdefenders through Project Glasswing. The capability ceiling is identical \u2014 what changes is who can access it and how strict the guardrails are.
Side by side
| Dimension | Claude Fable 5 | Claude Mythos 5 |
|---|---|---|
| Underlying weights | Same model | Same model |
| Safeguards | Strict safety classifiers | Lifted in some areas |
| Availability | Generally released (public, restored July 1, 2026) | Restricted \u2014 Project Glasswing only |
| Who can use it | Anyone | Approved US organizations (resumed July 1, 2026) |
| Sensitive-topic behavior | Under 5% of sessions quietly answered by Opus 4.8 | Answers directly within its remit |
| Benchmarks | Top scores on coding & knowledge work | Same, plus higher on restricted evals |
Why the scores look different
On the public table, Fable 5 leads on agentic coding (80.3% SWE-Bench Pro) and knowledge work. The gap shows up only on cybersecurity and biology evaluations like ExploitBench: there, Fable 5\u2019s safeguards cap its scores near Opus 4.8, while the published top numbers belong to the unrestricted Mythos 5. That is a guardrail artifact, not a difference in raw intelligence. See the full breakdown in our safeguard fallback explainer.
Which one are you using?
If you are a normal developer or app user, you are using Fable 5 \u2014 and on sensitive prompts, a small share of sessions are quietly handled by Opus 4.8 instead of being refused (slightly more common since the July 1 restoration, while Anthropic tunes the new safety classifiers). You are never silently upgraded to Mythos 5; that access is gated to approved US organizations under Project Glasswing. Fable 5 itself was suspended under US export controls from June 12 to July 1, 2026 and is now back worldwide \u2014 live status on Is Fable 5 back?.
So which should you care about?
- For your work, it is Fable 5. Mythos 5 is not something you can request; treat it as the same engine without your-side guardrails.
- If a sensitive task gets a cautious answer, that is the Opus 4.8 fallback inside Fable 5, working as designed \u2014 not a downgrade you can fix with a different model name.
- If a Fable 5 request gets rerouted, the practical question is not "can I get Mythos 5" (you cannot) but "what is the best alternative" \u2014 usually Opus 4.8, which is also the built-in fallback.
Bottom line
"Fable vs Mythos" is a question about access and safeguards, not capability. Same weights, two doors \u2014 and only one of them, Fable 5, is open to you (worldwide again since July 1, 2026). Keep a portable prompt and an alternative ready, and stop waiting on a Mythos 5 you cannot call.
Independent guide based on Anthropic\u2019s June 9, 2026 announcement. Live availability on our status page.