Atlas

Models

← All models

Gemini 2.5 Flash Preview 04-17

Google DeepMind · 2025-04-17

The April 17, 2025 preview snapshot of Gemini 2.5 Flash, kept separate from the later generally available release because provider behavior and benchmark results are snapshot-specific.

Benchmark scores

BenchmarkScore
Aider Polyglot47.1
AIME 202578.0
ARC-AGI-132.3
Artificial Analysis Intelligence Index17.5
Capability138.7
Confabulations Leaderboard Confabulation Rate4.5
Confabulations Leaderboard Non-Response Rate29.2
Confabulations Leaderboard Weighted Score16.8
CorpFin v254.2
CorpFin v2 - Exact Pages54.8
CorpFin v2 - Max Fitting Context52.3
CorpFin v2 - Shared Max Context55.4
CyberGym4.8
DeepResearchBench0.3
Epoch Capabilities Index (ECI)141
Fiction.liveBench47.2
ForecastBench Baseline Dataset Brier Index60.9
ForecastBench Baseline Market Brier Index60.4
ForecastBench Baseline Overall Brier Index60.5
ForecastBench Preliminary Dataset Brier Index60.9
ForecastBench Tournament Dataset Brier Index60.9
ForecastBench Tournament Market Brier Index67.5
ForecastBench Tournament Overall Brier Index64.0
GeoBench ACW Country Accuracy73.0
Global MMLU Lite88.4
GPQA Diamond64.6
GPQA Diamond (Vals protocol average)57.6
Humanity's Last Exam12.1
Humanity's Last Exam (Text Only)12.6
Humanity's Last Exam (Text-Only)11.6
Loading Atlas data…