Gemini 3.5 Flash-Lite

Google DeepMind's fastest and least-expensive model in the Gemini 3.5 family, released 2026-07-21 alongside Gemini 3.6 Flash. A low-latency, high-throughput reasoning model (thinking supported despite the Lite tier) built as a 'subagent' for high-volume automation -- agentic search, document processing, classification and routing. 1M-token context, natively multimodal (text/image/audio/video in, text out), ~350 output tokens/sec (official). Model ID gemini-3.5-flash-lite, generally available on the Gemini API/AI Studio/Vertex; $0.30/M input, $2.50/M output (cache-hit input ~$0.03/M). Benchmarks are Google self-reported from the joint launch blog, which shows large jumps over the prior Gemini 3.1 Flash-Lite (e.g. GDPval-AA v2 1140 vs 642, Terminal-Bench 2.1 54% vs 31%). No independent per-benchmark scores exist yet (Artificial Analysis publishes only a composite Intelligence Index of 36).