[AINews] Fable 5.2 retakes World #1 LLM spot as $ANT IPO roadshow begins
Fable 5.2 cuts token prices by 20%, but its heavier checking can raise the bill per task.
SemiAnalysis measured $88.15 per GDPx task, up from $67.16 for Fable 5.1, as the new model generated hundreds or thousands of tests and enlisted subagents to check its work. That helped it lead GDPx at 16.4%; it also topped Artificial Analysis’s Intelligence Index at 65. Effort settings matter: Anthropic says High matched Opus 5.5 at Max for roughly half the cost on FrontierCode 2.0, while several evaluations found higher settings could perform worse. The release coincided with the start of Anthropic’s IPO roadshow.
Latent Space's note
SemiAnalysis measured $88.15 per GDPx task, up from $67.16 for Fable 5.1, as the new model generated hundreds or thousands of tests and enlisted subagents to check its work. That helped it lead GDPx at 16.4%; it also topped Artificial Analysis’s Intelligence Index at 65. Effort settings matter: Anthropic says High matched Opus 5.5 at Max for roughly half the cost on FrontierCode 2.0, while several evaluations found higher settings could perform worse. The release coincided with the start of Anthropic’s IPO roadshow.
Latent Space's note
score 9