Atlas

Models

← All models

o3-pro

OpenAI · 2025-06-10

OpenAI o3-pro is a higher-compute version of the o3 reasoning model released on June 10, 2025, designed to deliver more thorough and consistently reliable answers on difficult problems. It is available in ChatGPT and through the Responses API, where it supports function calling, file and web search, image input, and other tools.

Benchmark scores

BenchmarkScore
Aider Polyglot84.9
ARC-AGI-157.0
ARC-AGI-24.9
Artificial Analysis Intelligence Index32.5
BuseyBench SVG4.0
Capability149.0
ClockBench Accuracy14.4
Confabulations Leaderboard Confabulation Rate23.4
Confabulations Leaderboard Non-Response Rate5.1
Confabulations Leaderboard Weighted Score14.2
Epoch Capabilities Index (ECI)148
Fiction.liveBench97.2
GPQA Diamond84.5
HardGeoBench <=100mi46.3
HardGeoBench <=10mi35.0
HardGeoBench Avg Distance1340
HardGeoBench Country Accuracy53.8
LLM Creative Story-Writing Benchmark8.4
LLM Writing Style Diversity0.2
MASK82.5
MCP Atlas Public44.5
MultiChallenge62.4
MultiNRC49.0
Professional Reasoning Benchmark - Finance49.1
Professional Reasoning Benchmark - Legal49.7
TutorBench54.6
VISTA51.6
WeirdML (v2)58.2
Loading Atlas data…