Google joins the decision-making model race: supports five modalities, with costs more than double that of Jev.

Dongcha Beating AI News Flash: Google has launched its first decision-making model, Gemini Decision 1, which supports mixed inputs of text, images, audio, video, and PDF, with a 256,000-token context window. Compared to Jev and OpenAI’s Decisions API, Google’s model can directly process a wider range of materials. For example, it can take a customer service recording, order details, and product photos together to simultaneously assess a customer’s request, issue severity, and need for human intervention. It can evaluate up to 1,000 queries per run; multiple queries sharing the same input material only incur a single input fee. The official end-to-end response time is 150 to 250 milliseconds. For inputs under 16,000 tokens, the model’s processing time remains nearly unchanged even when the number of queries increases from 1 to 100. In Google’s internal tests, Gemini Decision 1 scored 92.2% on JevBench, outperforming Jev 1.14 (90.0%) and OpenAI’s Decisions API (90.9%). However, it lagged Jev (82.0%) with a 79.9% score on the tool calling benchmark When2Call. In third-party evaluator Artificial Analysis’s Decision Index v2.1, Gemini Decision 1 posted an overall score of 65.98, ranking 9th among 18 models—nearly on par with Jev’s 66.23, but trailing top-ranked Drex Max v1.1’s 72.62. Google charges $0.1 per million input tokens, with free output. Calculated using Artificial Analysis’s standardized task set, completing 1 million tasks costs roughly $345 for Google, $156 for Jev, and about $103 for Drex Max. Google’s average task latency is 211 milliseconds, compared to 92 milliseconds for Jev. Google has deployed the new model for model routing in Gemini Notebook, reducing latency for some daily requests by up to 40%. The API is now available, with a free trial quota offered.