Google Launches Gemini Decisions API and Gemini Decision 1 With Audio, Video, and 1,000-Question Batches
Google entered the decision-model market on November 11 with a multimodal endpoint supporting text, images, audio, video, and PDFs at $0.10 per million input tokens.
Google became the third major hyperscaler to enter the decision-model market on November 11, launching the Gemini Decisions API powered by Gemini Decision 1—a non-generative model that returns calibrated yes/no, multiple-choice, and rubric scores with zero output tokens.
Broadest modality and batch scale in the category
Where early decision models focused primarily on text and later added static images, Google positioned Gemini Decision 1 on multimodal breadth and request scale. Developers can pass text, images, audio, video, and PDFs inside a 256K-token context window and evaluate up to 1,000 questions—with up to 255 choices per question—in a single request. Google lists input tokens at $0.10 per million with free output and reports end-to-end latencies of 150 to 250 ms.
How Gemini Decision 1 compares with Jev 1.14
In launch materials reported alongside SiliconANGLE’s coverage, Google said Gemini Decision 1 outperformed TypeSafe’s Jev 1.14 on five of seven shared internal benchmarks. Independent tracking from Artificial Analysis showed a closer contest: Gemini Decision 1 and Jev 1.14 tied at 66 on its Decision Index run, while Jev completed tasks noticeably faster (roughly 92 ms versus 211 ms) and at less than half the list input price ($0.042 versus $0.10 per million tokens).
Google also published a codelab featuring djev-run, an open server reimplementing Jev’s /v1/systemone interface on top of DiffusionGemma—further cementing TypeSafe’s request contract even as Google competes directly on hosted APIs.