Good morning!
So Mythos is finally here — and honestly, it's a bit wild.
Quick recap if you missed it: Mythos was this model Anthropic basically couldn't release, because it was too powerful for cybersecurity — it finds vulnerabilities in nearly every major piece of software out there. So they handed it to a bunch of top companies, got all the scary stuff patched, and now we've got a version we're actually allowed to use. It shipped as Claude Fable 5.
And of course, I had to point it at real code. So I did two things.
Fable 5 on my real SaaS (Scripti.ai)
Four tasks: find bugs, fix them, redesign the UI, run a security audit.
Result: 23 real bugs, an unprompted pricing-page redesign, and a critical security hole — anyone could self-grant my top paid tier with a shared promo code (plus credit-bypass race conditions, a leaking debug router, and rate-limit bypass).
The kicker: I'd run this exact pass before with Opus 4.7, 4.8, DeepSeek and Codex (GPT 5.5). None of them found these. Fable did — on code I hadn't touched in weeks.Fable 5 vs GPT 5.5 (Codex) — same app, same prompt
Fable won every round:
• Bugs: 23 found and fixed vs Codex's 1
• UI: a bold pricing redesign vs Codex's minor "polish"
• Security: a real exploit audit vs Codex just running npm audit (dependency advisories 13 → 0)
Codex isn't bad — it's tidy and disciplined. But on real, app-layer security it basically did nothing.
So look — I'll just say it: I'm going all back in on Claude.
Codex is cool, and I still use GPT for image gen, but for real code? Claude's just it. And Fable 5 is the icing on the cake.
Watch both:
→ Mythos / Fable 5 on real code: https://youtu.be/LoIGVdfTq9M
→ Fable 5 vs GPT 5.5 (Codex): https://youtu.be/YB2x8Mn4dUI
Listen — if you want to build and actually secure real apps with this stuff, that's exactly what we do inside the community. Come join us: https://www.skool.com/luke
Let me know what you think,
Luke
