Atlas

Models

← All models

Gemini 2.5 Pro

Google · 2025-06-05

Gemini 2.5 Pro is Google's multimodal reasoning checkpoint first released as gemini-2.5-pro-preview-06-05 on June 5, 2025, introducing configurable thinking budgets. Google promoted this same model unchanged to the stable gemini-2.5-pro endpoint on June 17, 2025. It supports multimodal input and a 1,048,576-token context window.

Benchmark scores

BenchmarkScore
AA-Briefcase Elo315
AA-Briefcase v1.1 Elo302
AA-LCR66.0
AA-LCR v1.169.0
AA-Omniscience Index-16.4
Active Probing Benchmark (APB)12.5
Active Probing Benchmark (Biology)7.5
Active Probing Benchmark (Chemistry)11.7
Active Probing Benchmark (Computer Science)8.3
Active Probing Benchmark (Economics)17.5
Active Probing Benchmark (Human Disease)26.7
Active Probing Benchmark (Mathematics)1.7
Active Probing Benchmark (Philosophy)13.3
Active Probing Benchmark (Physics)13.3
Agent Red Teaming k=100 (CAIS)91.4
Aider Polyglot83.1
AIME 202588.0
AIME 2025 (Artificial Analysis avg@10)87.7
AIME 2025 (MathArena repeated-run mean)88.3
ALE-Bench1175
AlgoTune1.5
Apex (MathArena)0.5
App-Bench0.0
ARC-AGI-1 Semi-Private41.0
ARC-AGI-24.9
Arena Code WebDev — Brand & Marketing1231
Arena Code WebDev — Consumer Product1221
Arena Code WebDev — Content Creation Tools1237
Arena Code WebDev — Data & Analytics1164
Arena Code WebDev — Frontend1225
Loading Atlas data…