Gemini Decision
by Google · Model · Language & code models
mixed
Multimodal yes/no model with confidence scores, but pricier and less accurate than rivals
Multimodal yes/no and multiple-choice model returning confidence scores; available in Gemini API with free tier
Why we say so
Artificial Analysis scores it 66 on its Decision Index, tied with Jev 1.14 and 7 points behind Drex Max and GLiDE. It also costs $345 per million tasks against $156 for Jev, and runs slower. Fahd Mirza found it good on text, images and PDFs but wrong on an audio test (distress labelled joy at 93%) and weak on spotting an AI-generated video. Google's claim of beating Jev on 5 of 7 benchmarks is not borne out by the independent index.
Evidence
- Nov 15On average scores better than Jev and other models on decision benchmarks
- Nov 13Gemini Decisions API launched: single-pass yes/no, multiple-choice (up to 255 options) and rubric scoring with confidence per answer
- Nov 12Text tests: blocked a data-deleting agent action at 99% and caught a prompt injection at 0.98; ticket routing was borderline (0.54 vs 0.40)
- Nov 12Per Google's table Jev still wins two benchmarks and is cheapest overall; Decision 1 costs about the same as GPT-6 Luna, far less than Flash-Lite
- Nov 12Scores 66 on the Decision Index, tying Elephant 9B and Jev 1.14 but 7 points behind Drex Max v1.1 and GLiDE v1.2
Show all 6
- Nov 12Google's table: leads most benchmarks against GPT-6 Luna and Gemini 3.5 Flash-Lite, with the lowest calibration error