Name: Gemini 2.5 Flash-Lite
Author: Google DeepMind

Gemini 2.5 Flash-Lite

Google DeepMind · 2025-07-22

Gemini 2.5 Flash-Lite is Google's cost-efficient, low-latency multimodal model, released as the stable gemini-2.5-flash-lite API version on July 22, 2025 after a June public preview. It accepts text, images, audio, and video, produces text, supports optional controllable thinking, and has a 1,048,576-token input limit.

Benchmark scores

Benchmark	Score
A-Fantasia	77.0
A-Fantasia Backwards Spelling	95.0
A-Fantasia Chess	60.0
A-Fantasia Cube Rotation	76.0
AA-LCR	51.3
AA-Omniscience Index	-44.6
Aider Polyglot	27.1
ALE-Bench	542
Artificial Analysis Intelligence Index	11.4
Artificial Analysis Omniscience Accuracy	17.6
Artificial Analysis Omniscience Hallucination Rate	75.4
BFCL v4 Overall Accuracy	36.9
BuseyBench SVG	2.6
Capability	133.9
CharXiv Reasoning	55.5
CritPt	0.0
FACTS Benchmark Suite	17.9
FormationEval	91.3
Global MMLU Lite	84.5
GPQA Diamond	66.7
Humanity's Last Exam	6.9
Humanity's Last Exam (Text-Only)	6.4
IFBench (Artificial Analysis)	49.9
Judgemark v2.1	56.0
LisanBench Difficulty-Weighted	38.2
LisanBench Path Length	122
LiveCodeBench	34.3
LiveCodeBench (Artificial Analysis)	59.3
MATH-500 (Artificial Analysis source)	96.9
MedCode	27.1