What happened
Latent Space’s AINews issue reviews Anthropic’s release of Claude Fable 5.2 and Mythos 5.2 alongside the start of Anthropic’s IPO roadshow. It reports Fable 5.2’s 16.4% score on SemiAnalysis GDPx, while the cost per task rose to $88.15 from $67.16 despite lower per-token prices. The issue attributes the benchmark result in part to more extensive test generation and subagent checks, and surveys other evaluations, access limits, and reactions.
Why it matters
The issue compares Fable 5.2’s benchmark gains with higher task costs, showing how extensive test generation and subagent checks can affect both performance and production economics.