AI Technology Observations | 13 November 2026, 07:00

This observation covers changes formally opened and verified in public material around the period from 7:51 am on 12 November to 7:51 am on 13 November 2026 in Australia/Sydney. It includes two items.

Google opens the Gemini Decisions API

On 11 November, Google opened the Gemini Decisions API, powered by the gemini-decision-1 model. It is designed for closed-set judgements. Rather than generating natural-language answers, one call can evaluate up to 1,000 independent yes-or-no, choice or rubric questions against the same text, image, audio, video or PDF input, returning probabilities and confidence scores. The developer documentation specifies an input limit of 262,144 tokens, while Google’s announcement rounds this to 256K. Choice questions can contain two to 255 options. Questions cannot read one another’s answers, so dependent decisions require a second call or application-side composition.

The new delivery is a dedicated decision endpoint with typed responses and no output-token billing, replacing the need to generate and parse text for this class of task. It is available through Google AI Studio and the Gemini API, with Python and JavaScript SDK version 3.2 or later. Standard pricing is US$0.10 per million input tokens and US$0 for output tokens, with a no-cost prototyping tier. Google reports 150–250 milliseconds of end-to-end latency and lower cost than its generative interface on a shared benchmark suite. Those latency, calibration and cost comparisons are developer measurements. VentureBeat independently confirmed the API’s availability, question count, supported inputs and price, and cited Artificial Analysis decision testing. There is not yet a public independent reproduction covering production traffic across multiple industries.

Verification used Google’s “Gemini Decisions API: Fast, calibrated multimodal decisions”, the Gemini Decisions API documentation, and VentureBeat’s independent report.

Google DeepMind releases the AlphaGenome 2 research API

Google DeepMind released AlphaGenome 2 on 11 November. The original AlphaGenome processed roughly one million DNA letters and predicted molecular readouts in parallel across multiple tissues and cell lines. AlphaGenome 2 is a conditional model: each run combines a DNA sequence of up to about one million letters with one tissue, cell type or cell-state condition, then predicts readouts such as gene expression, chromatin accessibility, histone modification and DNA methylation for that cellular context. The developer says the training data contained more than 107 million single-cell profiles across 68 tissues.

DeepMind reports that, on the TenK10K dataset covering 28 primary immune-cell types, the new model identified 30 per cent of causal variants compared with 16 per cent for the earlier model. It reports 43 per cent versus 23 per cent for the direction of gene effects. These figures come from the developer’s technical report and have not been fully reproduced independently. Nature confirmed the release and its research use in locating the cell types in which DNA variants act. Independent sources have not yet published a full reproduction of the developer benchmark.

AlphaGenome 2 is available through the AlphaGenome API for non-commercial research, while commercial research is in private preview. The announcement did not release model weights or state a date or complete price for public commercial access. The model is for research use only and has not been validated or approved for clinical or diagnostic use. Verification used the Google DeepMind announcement and technical report and Nature’s report.

Sources

Google, “Gemini Decisions API: Fast, calibrated multimodal decisions”, 11 November 2026, https://blog.google/innovation-and-ai/technology/developers-tools/gemini-decisions-api/

Google AI Studio, “Decisions API”, accessed 13 November 2026, https://aistudio.google.com/docs/decisions-api

VentureBeat, “Google’s Gemini Decision 1 model scores up to 1,000 questions in one API call”, 11 November 2026, https://venturebeat.com/technology/googles-gemini-decision-1-model-scores-up-to-1-000-questions-in-one-api-call

Google DeepMind, “AlphaGenome 2: Deciphering how cells read DNA to better understand disease”, 11 November 2026, https://deepmind.google/blog/alphagenome-2-deciphering-how-cells-read-dna-to-better-understand-disease/

Google DeepMind, “AlphaGenome 2” technical report, 11 November 2026, https://storage.googleapis.com/deepmind-media/DeepMind.com/Blog/alphagenome-2-deciphering-how-cells-read-dna-to-better-understand-disease/alphagenome-2.pdf

Nature, “DeepMind’s ultra-powerful AI for genome mapping wows scientists”, 11 November 2026, https://www.nature.com/articles/d41586-026-03673-0