Google has introduced Gemini Decision 1, a decision-model API that joins Jev and OpenAI in this emerging product category. Gemini Decision 1 supports the same three question types used by its competitors: yes/no, choice, and rubric-scoring. Google also reuses Jev's term "Noul" for yes/no questions.
Similarities and differences
- Question types: Gemini Decision 1 offers the same question formats as Jev and OpenAI decision models, making it functionally compatible for similar use cases. Jev has been limited to text and JSON inputs.
- Multimodality: whereas Jev accepts only text and JSON and OpenAI's decision models handle text and images, Gemini Decision 1 is fully multimodal and accepts text, images, audio, video, and PDF files as inputs.
Pricing
Gemini Decision 1 is priced the same as OpenAI's GPT-6 Luna decision model: $0.10 (10 cents) per million input tokens, with output tokens free. By comparison, Jev is currently cheaper at $0.042 (4.2 cents) per million input tokens.
Practical test: video input
The model was tested with a video file titled "Yonkers stomping robot". Four yes/no questions were submitted along with the video:
- "dragon": Does this video have a dragon in it?
- "robot": Does this video have a robot in it?
- "france": Is this video set in France?
- "usa": Is this video set in the USA?
In the reported test, Jev and OpenAI returned hard 1.0 or 0.0 values for yes/no questions. Gemini Decision 1 instead returned probabilistic yes/no outputs with less extreme confidence levels; the returned JSON included the following probabilities:
{ "dragon": { "type": "yes_no", "probability": 0.04976021841520312 }, "robot": { "type": "yes_no", "probability": 0.9234055387792485 }, "france": { "type": "yes_no", "probability": 0.08329171353166241 }, "usa": { "type": "yes_no", "probability": 0.9230459995157706 } }
These outputs indicate that Gemini Decision 1 assigned high probability to the presence of a robot and a USA setting in the video, and low probability to the presence of a dragon or a France setting.
Why this matters
The arrival of Gemini Decision 1 intensifies competition among decision-model offerings, especially around multimodal capabilities. Returning probabilistic answers rather than binary 0/1 labels can provide finer-grained information about model confidence, which may be valuable for downstream decision-making and error handling. Pricing and supported input formats will determine which service is more cost-effective for particular use cases.
Sources
The announcement and the test were discussed in community channels, including posts on Hacker News.
