Atlas

Models

← All models

o3

OpenAI · 2025-04-16

o3 is OpenAI's o-series reasoning model that uses extended chain-of-thought to tackle complex problems in math, science, coding, and visual reasoning. It was OpenAI's flagship reasoning model and supports tool use within its reasoning process. Its only published snapshot, and the alias's default, is o3-2025-04-16.

Benchmark scores

BenchmarkScore
AA-LCR69.3
AA-LCR v1.174.7
AA-Omniscience Index-15.6
Active Probing Benchmark (APB)25.3
Active Probing Benchmark (Biology)28.3
Active Probing Benchmark (Chemistry)21.7
Active Probing Benchmark (Computer Science)10.0
Active Probing Benchmark (Economics)35.0
Active Probing Benchmark (Human Disease)33.3
Active Probing Benchmark (Mathematics)20.0
Active Probing Benchmark (Philosophy)25.0
Active Probing Benchmark (Physics)29.2
Aider Polyglot76.9
AIME 2024 (avg@8)87.2
AIME 2024-202585.3
AIME 202588.9
AIME 2025 (Artificial Analysis avg@10)88.3
AIME 2025 (avg@8)83.3
AIME 2025 (MathArena repeated-run mean)89.2
ALE-Bench1304
APEX (Mercor)8.6
APEX-v1-extended60.5
ARC-AGI-1 Semi-Private60.8
ARC-AGI-26.5
Arena Text — Business, Management & Financial Ops (No Style Control)1393
Arena Text — Chinese (No Style Control)1437
Arena Text — Creative Writing (No Style Control)1358
Arena Text — English (No Style Control)1416
Arena Text — Entertainment, Sports & Media (No Style Control)1373
Arena Text — Exclude Ties (No Style Control)1393
Loading Atlas data…