Technology News and Headlines

Google Deepmind Gives Gemini 3 Flash the Ability to Actively Explore Images Through Code

Google DeepMind has introduced “Agentic Vision” to its Gemini 3 Flash model, enabling the system to actively explore images by generating and executing Python code. The model follows a think‑act‑observe loop—analyzing a request, planning actions, running code to zoom, crop or annotate, and feeding the results back into its context. This interactive approach yields a 5‑10% quality lift on standard vision benchmarks and has already helped a blueprint‑checking startup improve accuracy by 5%. The capability is currently limited to the Flash tier and still requires explicit prompts for many actions.