Anthropic's Fable 5: The world's most expensive, unavailable coding king
Anthropic just snatched the coding crown from OpenAI with Fable 5, but don't hold your breath. This pricey masterpiece beats GPT-5.5 by a hair, yet remains trapped behind a velvet rope of bureaucratic red tape.
The new DeepSWE leaderboard has a fresh leader, as Anthropic's Fable 5 hit 70% accuracy on complex engineering tasks. This benchmark is designed to be cheat-proof by using original problems rather than recycled code from GitHub repositories, which previously let models like Claude Opus scrape their way to higher scores by memorizing existing fixes.
However, the victory is razor-thin. Fable 5 outperformed GPT-5.5 by only three percentage points, yet costs roughly double to run. While the winner charges up to $21.63 per task, its rival delivers similar reliability for a modest $7.23. Statistically, the two models are basically neck-and-neck, with their confidence intervals overlapping so much that the title of 'best' is mostly a marketing flex.
The irony is peak Silicon Valley: even if you wanted to burn your budget on this marginal gain, you can't. Access to Fable 5 is currently suspended due to ongoing export control squabbles between Anthropic and US regulators. It turns out the most powerful coding assistant is currently just a very expensive ghost in the machine.
Paying a massive premium for a 3% boost that nobody can actually use is the ultimate tech-bro flex. It perfectly captures the industry's shift from building accessible tools to gatekeeping digital prestige, leaving users to wonder if these benchmarks are for engineers or just for venture capital pitch decks.
Source: DeepSWE Leaderboard
Comments
This is where the magic happens: AI reads your discussion and rewrites the article based on the most interesting comments. Each strong comment adds points to the meter below. Once the meter is full, the article updates live — no page reload needed.