Rankings / Google DeepMind

Gemini 2.5 Flash Lite Preview 09 2025

release date unknown

BenchAtlas Index

as of 2026-07-14

Not enough benchmark coverage for a composite score (needs ≥4 families across ≥3 categories). The per-benchmark evidence below stands on its own — sparse coverage is an evidence gap, not a low score.

Benchmark evidence

16 results

knowledge science

GPQA Diamond implementation=vals-ai	thinking	70.2 %	independent	Vals AI
GPQA Diamond implementation=vals-ai		64.1 %	independent	Vals AI
MMLU-Pro implementation=vals-ai	thinking	79.1 %	independent	Vals AI
MMLU-Pro implementation=vals-ai		78.6 %	independent	Vals AI

multimodal

MMMU implementation=vals-ai	thinking	75.4 %	independent	Vals AI
MMMU implementation=vals-ai		72.5 %	independent	Vals AI

professional

CorpFin	thinking	57.6 %	independent	Vals AI
CorpFin		56.3 %	independent	Vals AI
LegalBench	thinking	82.0 %	independent	Vals AI
LegalBench		79.0 %	independent	Vals AI
MedQA	thinking	88.9 %	independent	Vals AI
MedQA		80.3 %	independent	Vals AI
TaxEval		66.2 %	independent	Vals AI
TaxEval	thinking	64.7 %	independent	Vals AI

reasoning math

AIME implementation=vals-ai	thinking	42.1 %	independent	Vals AI
AIME implementation=vals-ai		26.3 %	independent	Vals AI