Claude Fable 5 outperforms GPT-5.6-terra in a physical AI evaluation, achieving a weighted score of 0.889 compared to 0.786 for GPT-5.6-terra, with a cost per trial of $9.60 and 52 graded runs in total. AI summary
Firehose
Filtered to Hacker News, tagged “physical” · clear filters
Browse: People · Companies · Papers · Podcasts · Hacker News · Deep dives