Google DeepMind announced AlphaGenome 2 on November 11. The DNA-sequence model predicts gene-regulatory effects in specified cell types and conditions, addressing information lost when different cells are averaged together.

Prediction, not a diagnosis

The model combines a long DNA sequence with a biological condition to predict experimental readouts, including gene expression and chromatin accessibility. Google reports improved research benchmark results; those evaluations do not demonstrate patient benefit or reliable diagnosis.

Non-commercial researchers can access it through the AlphaGenome API. Commercial research is in private preview, while a planned Google Antigravity integration is described as coming soon.

Keep the boundary clear

The model is for research and is not validated or approved for clinical or diagnostic use. It is also a specialized genomics system, not a new general-purpose Gemini model or a source of ordinary language-model token pricing.