Gemini 3.5 Flash-Lite

Name: Gemini 3.5 Flash-Lite
Author: Google

Google · 2026-07-21

Gemini 3.5 Flash-Lite is Google’s July 2026 cost-efficient model for high-throughput, latency-sensitive workloads such as subagents and document processing. It is based on Gemini 3.1 Flash-Lite, supports configurable thinking and multimodal input, exposes a 1,048,576-token input context through the Gemini API, and has a March 2026 knowledge cutoff.

Benchmark scores

Benchmark	Score
AA-Briefcase Elo	645
AA-Briefcase v1.1 Elo	651
AA-LCR	62.0
AA-LCR v1.1	76.7
AA-Omniscience Index	5.2
AA-Omniscience v1.1 Accuracy	26.5
AA-Omniscience v1.1 Hallucination Rate	26.7
AA-Omniscience v1.1 Index	6.9
AHK-Eval	55.6
AHK-Eval Algorithms	16.7
AHK-Eval Data Structures	50.0
AHK-Eval Date and Time	66.7
AHK-Eval Easy	75.0
AHK-Eval Hard	25.0
AHK-Eval Hidden Cases	60.2
AHK-Eval Mid	66.7
AHK-Eval Numbers	83.3
AHK-Eval Parse Success	86.1
AHK-Eval Regex	66.7
AHK-Eval Strings	50.0
AIME 2025	70.0
AIME 2026	82.2
ALE-Bench	1137
Antidote: Everyday Edition	991
APEX (Mercor)	7.8
APEX-Agents 1.1	46.1
APEX-Agents 1.1 Pass@1	29.3
APEX-SWE	31.2
APEX-v1-extended	54.7
ARC-AGI-1 Semi-Private	53.5