Anthropic’s Fable 5 entered the market on June 9 with a fanfare that few AI releases have seen. The model, billed as a Mythos‑class system, offered a one‑million‑token context window, 128,000 output tokens, and a performance edge that quickly made it the top‑ranked entry on the Chatbot Arena leaderboard. Within days, it was crushing OpenAI’s GPT‑5.5 on coding benchmarks, posting a 22‑point lead on SWE‑Bench Pro (80.3% vs. 58.6%) and a 95.0% score on the curated SWE‑Bench Verified subset. In the Code Arena, Fable 5’s Elo rating of 1,665 outpaced GPT‑5.5’s 1,501 by 98 points.

The advantage extended to the FrontierCode Diamond benchmark, where Fable 5 achieved 29.3% versus GPT‑5.5’s 5.7%. Even the broader Chatbot Arena placed Fable 5 at number one, pushing GPT‑5.5 down to fourth. The only area where GPT‑5.5 narrowed the gap was Terminal‑Bench 2.0, a test of live terminal‑based coding tasks, where it scored 82.7% against Fable 5’s roughly 88%.

Pricing, however, tilted the scales toward OpenAI. Developers could run GPT‑5.5 for $5 per million input tokens and $30 per million output tokens—half the cost of Fable 5’s $10 and $50 rates. For high‑volume applications where cost outweighs marginal performance gains, the cheaper model remained the pragmatic choice.

The rapid ascent of Fable 5 was cut short on June 12 when the U.S. Department of Commerce issued an export‑control directive, citing a jailbreak vulnerability. The order forced Anthropic to take both Fable 5 and the broader Mythos‑5 family offline. Anthropic argued the vulnerability was minor, already public, and exploitable in GPT‑5.5 without any special bypass techniques. Internal reports suggest Amazon CEO Andy Jassy played a role in prompting the government review.

For developers who had begun evaluating Fable 5 for production workloads, the shutdown meant a sudden pivot back to GPT‑5.5 or Anthropic’s older Opus models. The performance downgrade is stark: the 22‑point SWE‑Bench Pro gap translates to a model that resolves four out of five real‑world software issues versus one that handles roughly three out of five.

Anthropic has opened negotiations with the Commerce Department, maintaining that the export‑control classification is disproportionate. Until a resolution is reached, GPT‑5.5 retains the top spot in the market—not because it is the best model that exists, but because its only serious competitor has been removed.

Este artículo fue escrito con la asistencia de IA.
News Factory APP - noticias agénticas para impulsar tu SEO y AEO.