Gemini Decisions API

Commenters discussed Google’s Gemini Decisions API as a decision-focused model and competitor inspired by Jev, with some wondering why the approach had not become a product category sooner. One commenter said its effectiveness is surprising given how simple the system seems, and argued that decision models challenge assumptions about intelligence. Others compared it with Google’s Embedding Gemma decision-making guide, asked when the API would reach Vertex, and noted that it is currently available through AI Studio. A commenter judged it more expensive than similar models but potentially valuable for decisions based on audio or video. Another cited benchmark results that appeared to favor Google’s model over OpenAI’s and Jev’s, while others questioned how much confidence to place in those comparisons.

The technical discussion focused on how models produce decision probabilities. One commenter suggested extracting log probabilities for “yes” and “no” variants, while another cautioned that fine-tuning and reinforcement learning can make those probabilities diverge from ground truth. Commenters also speculated about the competitive implications: one wondered whether Jev could survive larger companies entering the field, while another noted Jev’s substantial recent financing. Views on Google’s commitment were divided. Some argued the API is a model product, not an app likely to be discontinued, while others pointed to Google’s history of shutting products down and expressed reluctance to rely on its newer services. One commenter also criticized Google’s handling of Jules, treating it as evidence that users may not get a clear migration path when products change.