Models RSS

Models 🇷🇺

Claude Fable 5 Outscores GPT-5.6 in From-Scratch Code Rewriting. Here's Why

Epoch AI's MirrorCode benchmark shows Claude Fable 5 solving 64% of tasks versus GPT-5.6 Sol's 20%. The gap stems from reliability, not raw capability: Sol often solves tasks partially, while Fable consistently completes them. Fable's strong performance in Ada, a language with scarce training data, suggests skill rather than memorization.

AnthropicAnthropic OpenAIOpenAI
Habr — хаб ИИ03.08 · 23:02
Fresh news