Atlas

Models

← All models

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

BenchmarkScore
A-Fantasia77.0
A-Fantasia Backwards Spelling95.0
A-Fantasia Chess60.0
A-Fantasia Cube Rotation76.0
AA-LCR51.3
AA-Omniscience Index-44.6
Aider Polyglot27.1
AIME 202563.1
AIME 2025 (Artificial Analysis avg@10)53.3
ALE-Bench542
Artificial Analysis Intelligence Index11.4
Artificial Analysis Omniscience Accuracy17.6
Artificial Analysis Omniscience Hallucination Rate75.4
BFCL v4 Overall Accuracy36.9
BuseyBench SVG2.6
Capability133.8
CharXiv Reasoning55.5
CritPt (avg@5)0.0
FACTS Benchmark Suite17.9
FormationEval91.3
Global MMLU Lite84.5
GPQA Diamond66.7
GPQA Diamond (Artificial Analysis avg@5)62.5
Humanity's Last Exam6.9
Humanity's Last Exam (Text-Only)6.4
IFBench (Artificial Analysis)49.9
Judgemark v2.156.0
LisanBench Difficulty-Weighted38.2
LisanBench Path Length122
LiveCodeBench34.3
Loading Atlas data…