Gemini 3.5 Flash-Lite

Gemini 3.5 Flash-Lite is not a good review seat: it caught 11% [3%–21%] of the known bugs and ranks 16 of 17, with 15 ranked models measurably ahead of it and 0 measurably behind. Pick something else.

Its price does not rescue it. A review cost $0.0059, which is 2.1× a GPT-6 Luna review and 1.5× a Claude Haiku 5.5 review, and both measurably beat it on recall. The list rate ($0.30 / $2.50 per 1M tokens) is low, but list rate is not the measured bill. If you want a cheap first seat, Luna is the better choice on this data.

Its output is thin: it wrote very few output tokens per review, so its reviews are short. Its noise is low too: 0.4 [0.2–0.7] false alarms per clean PR. A model that says little is quiet on clean code and on buggy code alike, so a low false-alarm count means little next to a recall this low.

Its misses include "Add weight option" and "Fix cache file length"; the bug table shows what it did catch. The per-language and per-kind slices are too small to say where it does better.

Compare it with Luna or read how to choose a model.