Claude Fable 5.1 takes InferenceBench lead, a user reports
A user reports a 9.83× geometric-mean speedup over a naive PyTorch solution. Another cites a 7.90× speedup for GPT-6-Astra (Ultra).
TLDR
A user reports that Claude Fable 5.1 has overtaken previous leader Opus 5 on InferenceBench, reaching a 9.83× geometric-mean speedup over a naive PyTorch solution. They suggest the jump mostly reflects stronger reasoning rather than new strategies. A second user quoting the result also describes Fable 5.1 as the leader and cites a 7.90× speedup for GPT-6-Astra (Ultra).
Combined views
14K
3 Sources, first seen 20d ago