Anthropic’s Fable 5 briefly outperformed OpenAI’s GPT 5.5 across major AI benchmarks before a June 12 U.S. export control directive took it offline. Key Points: Fable 5 led GPT 5.5 on Chatbot
Anthropic’s Fable 5 briefly outperformed OpenAI’s GPT 5.5 across major AI benchmarks before a June 12 U.S. export control directive took it offline.
Key Points:
- Fable 5 led GPT 5.5 on Chatbot Arena, SWE-Bench Pro and major coding tests.
- The model was available for only three days before the U.S. government ordered Anthropic to disable it.
- GPT 5.5 is now the strongest available model by default, not because it passed Fable 5.
Fable 5 became the most capable public AI model after its Jun. 9 launch, topping GPT 5.5 on major benchmarks before the U.S. government intervened three days later.
The model ranked first on Chatbot Arena, while GPT 5.5 ranked fourth. On SWE-Bench Pro, Fable 5 scored 80.3%, compared with 58.6% for GPT 5.5, a gap of nearly 22 points in real software engineering tasks.
The lead was also clear in coding tests. Fable 5 scored 1,665 on Code Arena, 98 Elo points above GPT 5.5, and reached 29.3% on FrontierCode Diamond, where GPT 5.5 managed 5.7%.
GPT 5.5 held one narrower advantage in practical positioning. It costs $5 per million input tokens and $30 per million output tokens, while Fable 5 cost $10 and $50, making OpenAI’s model cheaper for high-volume use.
Fable 5 also offered a one-million-token context window and 128,000 output tokens. Anthropic had made it available to Pro, Max, Team and Enterprise subscribers at no extra cost until June 22, before the order ended that window early.
Also Read:Is AI Becoming A Real Advantage In Court? Ask The Lawyer Who Just Beat Meta
The shutdown followed a Jun. 12 export control directive that cited a jailbreak vulnerability in Fable 5 and the wider Mythos 5 model family. Anthropic disputed the finding, saying the issue was minor, already known and also achievable on GPT 5.5 without special bypass methods.
The result is unusual for the AI market.
Developers lost access to the model that led the benchmark tables, while GPT 5.5 became the best available option because its closest rival was removed.
That distinction matters most for coding workflows. A 22-point SWE-Bench Pro gap means the difference between a model that can solve about four in five real codebase issues and one that handles closer to three in five.
Fable 5’s brief run also showed how fast the frontier can move. GPT 5.5 launched in late April under the internal codename “Spud,” but its lead lasted only until Anthropic opened public access to a stronger Mythos-class system in June.
Read Next:Anthropic Refused To Patch Claude Fable's Jailbreak, So The US Banned It, David Sacks Says
