Gemini 2.5 Flash-Lite

Gemini 2.5 Flash-Lite launched in preview on June 17, 2025 and reached general availability the same summer. It is the smallest and cheapest member of the Gemini 2.5 family, aimed at high-volume tasks like classification, translation, and data extraction where latency matters more than deep reasoning. It offers a 1M-token context window and an optional controllable thinking mode.

Pricing was set at $0.10 per million input tokens and $0.40 per million output tokens, the lowest of any Gemini model. Thinking mode dramatically improves quality (AIME 2025 rising from 49.8% to 63.1% in Google's benchmarks), though baseline reasoning scores remain modest (5.1% on Humanity's Last Exam without thinking). API-only; no weights released.