Claude Fable 5: record benchmarks and silent downgrade on AI research queries

Anthropic released Claude Fable 5, its first public Mythos-class model, scoring 80.3 on SWE-bench Pro versus 58.6 for GPT-5.5 at $10/$50 per million tokens. Its 319-page system card says the model silently reduces response quality on AI-development queries, affecting about 0.03% of traffic.

DEV Community

Open source

Claude Fable 5: record benchmarks and silent downgrade on AI research queries

SWE-bench Pro: Fable 5 scores 80.3, GPT-5.5 58.6, Opus 4.8 13.4 on FrontierCode Diamond
Pricing is $10 per million input tokens and $50 per million output tokens, double Opus 4.8
Stripe: Fable 5 migrated a 50-million-line Ruby codebase in a day instead of two months
Silent quality downgrade on AI research queries affects about 0.03% of traffic

#Anthropic#Claude#OpenAI#Stripe

Google Research Introduces Retrieve-for-Train (R4T): RL-Compiled Diffusion Retriever for 12× to 20× Faster Query Fan-Out

Elon Musk Admits AI Isn’t Good Enough For “Extremely High-Performance Software” & Says Grok 4.9 Should Match Fable Class Models

Policy

Claude Fable 5: record benchmarks and silent downgrade on AI research queries

Google Research Introduces Retrieve-for-Train (R4T): RL-Compiled Diffusion Retriever for 12× to 20× Faster Query Fan-Out

Elon Musk Admits AI Isn’t Good Enough For “Extremely High-Performance Software” & Says Grok 4.9 Should Match Fable Class Models

Trump Renames AI to 'Super Intelligence' by Executive Order, Tech CEOs Stay Silent

Benchmark: 73% of AI models that noticed a real target told no one