Google Gemini Decision 1: Scoring 1,000 Questions in One Pass

What happened

Google DeepMind has launched the Gemini Decisions API, with Gemini Decision 1 as its first model. The system evaluates up to 1,000 closed-ended questions in a single pass, and it accepts text, images, audio and video inputs. The launch was reported by The Decoder.

According to Google, Gemini Decision 1 returns results in roughly 150 to 250 milliseconds. Google also says it costs around five times less to run than Gemini 3.5 Flash-Lite. The Decoder frames the release as part of a wider wave of decision-focused models, although the report covers only this launch and its published claims.

The core proposition is consolidation. Work that would normally require many separate queries, or a chain of model calls, can be batched into one evaluation. Because the reporting is based on Google's own figures, the speed and cost comparisons have not yet been independently verified.

Why it matters

For leaders in AI and digital transformation, the significant change is what this type of model makes practical. A system that can score a thousand structured questions in a fraction of a second, across several media types, can move from occasional use in a pilot to a routine part of operations. Eligibility checks, quality reviews, compliance screening and triage are all tasks that are naturally framed as closed-ended questions. Lower latency and cost make them easier to run continuously rather than in periodic batches.

The operating-model question follows from that. When decisions can be evaluated at this scale and speed, the constraint shifts from computing capacity to how well an organisation defines its questions, its criteria and its escalation rules. Teams that treat these decisions as a designed service, with clear standards and human review points, are likely to get more value than those that simply add the model to existing workflows.

By the numbers

• 1,000 closed-ended questions can be evaluated in a single pass, according to the launch report.
• 150 to 250 milliseconds is the response time Google gives for Gemini Decision 1.
• About five times lower cost than Gemini 3.5 Flash-Lite, according to Google.

The Renascence take

Most coverage will focus on the headline number of 1,000 questions. The more useful question is how many of your current decisions are already written as rules, forms or checklists that a model could evaluate in one go, because those are the processes where this kind of speed changes the service.

The real design challenge is not the model, it is the question set. Speed and low cost make it easy to automate poorly framed decisions at scale, which means a customer could be judged quickly and still unfairly. Customer-obsessed operators should audit their decision criteria before they expand them, identify the points where a customer needs a clear reason and a route to appeal, and keep a human accountable for those moments.