Google DeepMind has unveiled Gemini 3.2, the latest iteration of its flagship AI model, featuring native real-time video understanding as its headline capability. Unlike previous models that processed video frame-by-frame, Gemini 3.2 can reason across continuous video streams, enabling applications like live sports analysis, real-time security monitoring, and interactive video tutoring. The model processes up to 4K resolution at 30 frames per second with what Google describes as “near-zero latency.”
Early benchmarks show Gemini 3.2 scoring 87.3% on the new VideoQA-Live benchmark, significantly ahead of GPT-5.2’s 71.8% and Claude Opus 4.6’s 68.4% on the same test. However, on traditional text and code benchmarks, the improvements over Gemini 3.1 Pro are marginal. Google is rolling out Gemini 3.2 to Vertex AI enterprise customers first, with broader API access expected by mid-April 2026.