Google Launches Gemini Decisions: 1,000 Questions per Query

Google launched Gemini Decisions on November 11, an API associated with the Gemini Decision 1 model that accepts up to 1,000 independent questions on the same input, within the limits of the available context. It processes images, audio, and video, but the comparative results available only pertain to text-based tasks.
Jev Equals Gemini in Textual Challenges
The Artificial Intelligence Decision Index v2.1 assigns a score of 66 to both Gemini Decision 1 and Jev 1.14. Its six evaluations examine document decisions, tool selection, knowledge, reasoning, and coherence in the face of reformulations. They measure both the accuracy and the confidence expressed by the models. Since the exercises are textual, these results do not constitute an assessment of visual, audio, or video capabilities.
For Gemini, tested from Google AI Studio, the index indicates 211 ms per task and estimates a cost of $345 for one million tasks. Microsoft Decision 1, accessible here via OpenRouter, scores 62, with 274 ms and an estimated cost of $143. Jev 1.14, used at TypeSafe, reaches 66 with 92 ms and $156, while GPT-6 Luna Decisions, from OpenAI, shows a score of 60 with 206 ms and $339. In these exercises, Jev therefore requires less than half the time and cost of Gemini, at an equal score.
These scores are rounded, and the indicated duration results from a combination of medians according to the index's weightings. The financial estimates apply the displayed rates to the benchmark tasks.
Le brief IA que les pros lisent chaque soir
Les 7 actus IA du jour, décryptées en 5 min. Gratuit.
Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.
Choisissez votre rythme
Gratuit · Pas de spam · Désabonnement en 1 clic
Google Also Accepts Recordings as Evidence
Google's service associates evidence with a fixed question: it can quantify the probability of a condition, select a provided option, or score an item according to a rubric. The software that receives these responses can use them to guide a request or submit it for review. In addition to recordings and images, accepted data includes texts and PDFs.
For testing, Google AI Studio has a dedicated space and a free plan intended for prototyping. The decision model can be experimented with alongside other Gemini models. The paid offer charges $0.10 for one million incoming tokens.
Formats and Pricing of Competing Offers
At Microsoft, Decision 1 receives text or JSON through Foundry, which also hosts models from OpenAI and Anthropic. Its displayed price is $0.042 for one million incoming tokens, the same as Jev, offered by TypeSafe. The latter now also accepts images on an experimental basis; its Jev 1.14 version can delineate a visual target with a bounding box. Jev and Gemini Decision 1 are also accessible via OpenRouter.
OpenAI's Decisions offering is generally available with GPT-6 Luna and accepts text, images, or JSON, with a cap of 400 questions per request. An image can be used for the entire request or attached to a specific question. Instruction caching is supported; outside of cache, the announced rate is $0.10 for one million incoming tokens, before applicable adjustments. None of the four offers charge for tokens produced as output.
Brief IA — L'actualité IA en français
L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.