Google releases Gemini Decision 1, saying it wins 5 of 7 shared benchmarks against Jev
Google 杀入 Jev 核心赛道:Gemini Decision 1 五项基准领先两项落后
AI summary
Google releases the multimodal decision model Gemini Decision 1 to challenge TypeSafe Jev, winning 5 of the 7 shared benchmarks published by Google. The model does not generate text; it directly outputs candidate probabilities and confidence, supporting up to 1,000 independent judgments per request. However, it charges 0.10 dollars per 1 million input tokens, about 2.38 times that of Jev, and its composite score on third-party leaderboards is also slightly lower than Jev's.
Read original