Gemini 3 Flash

Gemini 3 Flash is a multimodal large language model from Google DeepMind, released on February 27, 2026, as the direct successor to Gemini 2.5 Flash in the company's fast-inference model line. On the MMLU-Pro knowledge benchmark (5-shot, CoT prompting), it achieved a score of 88.6, an increase of 3.2 points over Gemini 2.5 Flash. Its Chatbot Arena Elo rating, sampled on March 1, 2026, stood at 1473, reflecting a statistically significant lead over its predecessor's final recorded Elo of 1421. The model is priced at $0.50 per million input tokens and $3.00 per million output tokens, matching its forerunner's output cost while reducing input cost by 33%. It processes text and image inputs natively, with a 1-million-token context window. Gemini 3 Flash matters now because its March 2026 Arena snapshot and specific MMLU-Pro prompting protocol provide a reproducible performance-cost anchor. This allows developers to directly measure the step-change between second- and third-generation Flash models, and to benchmark Google's lightweight offering against contemporaneous fast-inference models like Claude Opus 4 and GPT-5.1 Turbo.