Name: Gemini 3.1 Flash-Lite
Author: Google DeepMind

Gemini 3.1 Flash-Lite

Google DeepMind · 2026-05-07

Gemini 3.1 Flash-Lite is Google's stable, low-latency Gemini 3 model for high-volume and cost-sensitive workloads. Google released the stable endpoint in May 2026 as the replacement for the retired Gemini 3.1 Flash-Lite Preview endpoint; it supports multimodal input, text output, function calling, code execution, and configurable thinking.

Benchmark scores

Benchmark	Score
A-Fantasia	53.3
A-Fantasia Backwards Spelling	74.0
A-Fantasia Chess	26.0
A-Fantasia Cube Rotation	60.0
ALE-Bench	798
BLXBench	64.2
BLXBench Coding category	94.3
BLXBench Cost category	88.0
BLXBench Debugging category	68.1
BLXBench Hallucination category	66.9
BLXBench pass rate	43.8
BLXBench Reasoning category	49.9
BLXBench Refactoring category	43.8
BLXBench Security category	61.3
BLXBench Speed category	97.2
BLXBench UI category	34.8
BuseyBench SVG	3.1
CadQueryEval	36.0
CadQueryEval — Bounding Box	56.0
CadQueryEval — Chamfer Distance	48.0
CadQueryEval — Code Execution	60.0
CadQueryEval — Component Count	52.0
CadQueryEval — Hausdorff 95p	40.0
CadQueryEval — STL Creation	60.0
CadQueryEval — Volume	48.0
CadQueryEval — Watertight Mesh	52.0
Capability	144.3
CharXiv Reasoning	73.2
CharXiv Reasoning (with tools)	75.6
ChessBench Accuracy	64.6