I had 3 days with it. That’s all anyone got.
Claude Fable 5 launched June 9. I immediately threw it at something non-trivial: a Figma clone. Not a toy. A real canvas editor with layers panel, design panel, multi-frame support, infinite canvas, and undo history.
Here’s what /cost showed after the session:
Total cost: $9.54
Total duration (wall): 28m 29s
Total code changes: 2889 lines added, 7 lines removed
claude-fable-5: 6.8k input, 93.8k output, 2.4m cache read
Enter fullscreen mode Exit fullscreen mode
Ctrl-Z worked. 22 smoke tests passed. Headless rendering was compliant.
For context: this is the kind of thing a junior dev would spend several days on. Fable did it in one session, with a simple prompt, without losing architectural coherence.
Then on June 12 at 5:21 PM ET, the US government issued an export control directive. Anthropic pulled the model immediately, for everyone, worldwide, no explanation given.
Here’s what I find hard to ignore.
Anthropic spent months documenting why Fable was safe for general use. Defense-in-depth classifiers. Opus 4.8 as a safety fallback on high-risk domains. No universal jailbreak found in testing. They published the evidence.
OpenAI filed its S-1 confidentially on May 22. IPO targeted for Q4 2026. The US government holds a stake.
Fable 5 was, by most benchmarks, the most capable model publicly available. It was suspended 3 days after launch.
I’m not drawing a direct line between those facts. I’m just saying the timing is what it is.
Anthropic did the safety work seriously. They got cut anyway.
OpenAI is losing $1.22 for every $1 of revenue, heading to an IPO, and ChatGPT keeps running fine.
This is America.
I’ll update this post when Fable comes back. My bet: shortly after the OpenAI IPO closes.
답글 남기기