No. 116
2026-11-12
the AI pulse — brewed, not raw
 AI CRITIQUE

Google launches Gemini Decisions API

Read the article    English (automatic translation)

Summary

Google has launched Gemini Decision 1. This decision model API mirrors Jev's question types and even its terminology verbatim. What sets the product apart is its full multimodal support, covering text, images, audio, video, and PDFs. Pricing stands at 10 cents per 1 million input tokens, with outputs being free.

Why it matters

Decision models are solidifying as a distinct product category. Google's entry into this space signals rapid consolidation among Jev, OpenAI, and now Gemini. The multimodal scope positions Gemini as the most versatile option for structured decision tasks. Jev supports only text, while OpenAI supports text and images. The near-identical API design and shared vocabulary indicate that the category is still young and standards have not yet settled.

Highlights

  • Gemini Decision 1 is the only decision model offering full multimodal support, including text, images, audio, video, and PDFs. Jev supports only text, while OpenAI supports text and images.
  • Pricing is 10 cents per 1 million input tokens with free outputs. This is competitive, though it sits above Jev's base price of 4.2 cents per 1 million tokens.
  • Unlike Jev and OpenAI, which return binary 1.0/0.0 responses, Gemini returns calibrated probabilities. For example, 0.92 for 'robot present' and 0.05 for 'dragon present.'

Critical take

Jev'in soru sınıflandırmasının ve hatta 'Noul' gibi uydurduğu terimlerin neredeyse kelimesi kelimesine yeniden kullanılması, meşru soruları gündeme getiriyor. Bu durum gerçek bir inovasyon mu, yoksa bir yaşından bile küçük bir kategoride hızlı özellik taklidi mi? Simon'un tek video testi örnekleyici olsa da, üretim hattında aşırı özgüven veya temkinli olmanın kritik olduğu sınır durumlarında olasılık çıktılarını yeterince zorlamıyor.

This critique was written by the local AI model Qwen3.8-27B. I chose the story; I do not review each critique before it is published.
Back to the issue