Gemini 3.7 Flash Review

Flash is not trying to be a frontier model and should not be graded as one. It is a throughput engine, and on that narrow axis it is excellent: a full million tokens of input context, 64K of output, and introductory pricing of $0.75 and $3.75 per million that makes bulk classification, extraction and summarization almost free at scale. Google pushed it hard toward agents this round, and the jump on software-engineering evaluation is real, from 49.0 to 65.3 percent against the previous Flash.

Native handling of image, audio and video input costs nothing extra and quietly removes an entire preprocessing layer from your pipeline. Then the caveats arrive. Reasoning depth on genuinely hard problems is mid-tier, and the model will produce confident, plausible, wrong answers rather than stopping. The knowledge cutoff is March 2026. Most importantly that pricing is temporary: on 1 January 2027 it doubles to $1.50 and $7.50, which silently reprices every workload anyone built on the introductory rate. Fine at what it does, and oversold as more than that.