Anthropic has released Claude Fable 5. The first publicly available Mythos-class model. Dropped today. Free on your plan until June 22.
Everyone is talking about "the most powerful model ever." Nobody is reading what it actually does.
Here are 10 things I found in the official documentation.
1. It Beat Pokémon FireRed. Using Only Its Eyes.
Not a joke.
Previous Claude models struggled to play Pokémon FireRed even with helper harnesses — maps, navigation aids, game-state information.
Fable 5 completed the game from start to finish using only raw game screenshots. No maps. No navigation. Just what it sees on screen.
This isn't a demo trick. It's a signal of how much vision improved. Fable 5 is now the state-of-the-art model for vision tasks.
2. It Migrated a 50-Million-Line Codebase in One Day.
Stripe gave Fable 5 their Ruby codebase. 50 million lines.
Task: codebase-wide migration.
Normally takes a whole team over two months by hand.
Fable 5 did it in one day.
This isn't "helped with code." This is autonomous engineering work that previously required a team.
3. It Has Memory That Works 3x Better Than Previous Models.
Fable 5 can use persistent file-based memory during long-running tasks.
Test: Slay the Spire (a deck-building game requiring strategy and memory).
Result: access to persistent memory improved Fable 5's performance three times more than it improved Opus 4.8's. Fable also reached the game's final act three times more often.
In practice: on long projects, Fable 5 takes its own notes and uses them to improve its outputs over time.
4. It Developed Drug Candidates for 9 out of 14 Protein Targets.
This is from the official announcement. Not marketing.
Mythos 5 (the full version of Fable 5 without safety filters) accelerated aspects of the drug design process by around ten times.
Without human assistance, the model:
- Chose binding sites
- Selected and ran protein design tools
- Recovered from failures along the way
9 of the 14 protein targets yielded strong drug design candidates currently under investigation.
One Mythos hypothesis — a novel mechanism for an E. coli protein — was independently corroborated in a study from a lab working on the same problem without knowing about Mythos.
5. It Simulated the Solar System From Physics First Principles.
And used it to predict solar eclipses.
Fable 5 didn't use ready-made libraries. It derived the planets' orbital motion from physics first principles and built a working simulation.
This is a level of physics understanding most people haven't processed yet.
6. It Played Factorio Autonomously. Built a Full Factory From Scratch.
Factorio is a factory-building game beloved by engineers.
Fable 5 played it strategically and built an automated factory on its own.
Why this matters: Factorio requires long-term planning, resource management, and understanding complex systems. This isn't random moves — it's real engineering strategy.
7. It Built a 3D-Printable Model in a CAD Editor. That It Also Built.
Fable 5 designed a complete 3D-printable model in a browser-based CAD editor.
The CAD editor was also created by Fable 5.
Including the built-in AI copilot that does the modeling.
The tool + the output of the tool. In one session.
8. It Wrote Music. In Code. Having Never Heard Music.
A fluid simulation where the motion is synchronized to the beat of a classical music EDM remix.
The remix was created by Fable 5 using code.
The model has never "heard" music in any conventional sense. It understood musical structure through patterns and reproduced it programmatically.
9. It Beat a Model Published in Science Journal. Being 100x Smaller.
Mythos 5 conducted novel genomics research over more than a week of largely autonomous work.
It assembled single-cell data for millions of cells spanning 138 animal species.
It designed and trained a custom machine learning model.
Result: outperformed a model published in the journal Science — while being 100 times smaller in size.
With only high-level human input.
10. It Can't Be Jailbroken. 1000+ Hours of Attempts. Zero Successes.
Before release, Anthropic ran an internal bug bounty trying to break Fable 5.
Over 1,000 hours of testing.
Zero universal jailbreaks found.
Then they brought in external red-teaming organizations. Also zero.
This is the first publicly available Mythos-class model to pass this level of security testing.
Test It Yourself — 10 Prompts That Show What Fable 5 Actually Does
Don't take my word for it. Run these yourself.
Test #1 — Vision (Feature 1)
Take any screenshot. Upload it to Fable 5 and send:
Test #2 — Long autonomous coding (Feature 2)
Paste 3–5 files from any codebase and send:
Test #3 — Memory and notes (Feature 3)
Start a long session and send first:
Test #4 — Scientific reasoning (Feature 5)
Test #5 — Build a tool that builds something (Feature 7)
Test #6 — Music/pattern synthesis (Feature 8)
Test #7 — Research synthesis (Feature 9)
Paste 3–5 research papers or long articles and send:
Test #8 — Devil's advocate (Feature 10)
Test #9 — Strategy and systems thinking (Feature 6)
Test #10 — The long-context stress test
This one specifically tests what separates Fable 5 from everything else:
How to Try It Right Now
Claude.ai — Fable 5 is live in the apps. Select it from the model dropdown.
API — model string: claude-fable-5
Amazon Bedrock — available now.
Free window: included on Pro, Max, Team, and Enterprise plans until June 22, 2026.
After June 22, usage credits may be required until capacity expands.
You have 13 days of free access to the most powerful publicly available AI model ever built.
The One Detail Everyone Is Missing
GitHub's changelog notes that Claude Fable 5 requires up to 30 days of prompt-and-output retention to operate the safety classifiers.
Your prompts are being retained for 30 days for safety monitoring.
That's the price of Mythos-class capability in public access.
Know this before uploading confidential data.
Conclusion
Anthropic spent two months saying Mythos was too dangerous.
Today they released it to anyone with a credit card.
The safeguards are real. The capability is also real.
The 10 features above aren't marketing. They're from official documentation and real partner testing.
Try it before June 22 while it's free.
*Educational content. Always review AI output before deploying to production.*
Links
- My Telegram: https://t.me/kirillk_web3
- My Twitter/X: https://x.com/kirillk_web3
- Hosting for Claude: https://ishosting.com/
