Atlas

Models

← All models

GPT-5 Pro

OpenAI · 2025-10-06

GPT-5 Pro is the extended-reasoning tier of OpenAI's GPT-5 generation, which spends more compute on deeper, more reliable reasoning for the most demanding professional and technical work. It targets the hardest problems in domains like science, math, and coding.

Benchmark scores

BenchmarkScore
ARC-AGI-170.2
Capability153.3
ClockBench Accuracy11.1
DeepSearchQA (F1 score)79.0
Epoch Capabilities Index (ECI)150
FORTRESS Average Risk Score (ARS)15.2
FrontierMath Tier 4 (v2) - Private19.5
FrontierMath Tiers 1-3 (v2) - Private55.8
Humanity's Last Exam31.6
Humanity's Last Exam (Text Only)33.3
LLM Writing Style Diversity0.2
MASK86.0
Miklós Schweitzer 202590.0
MultiNRC65.2
Professional Reasoning Benchmark - Finance51.1
Professional Reasoning Benchmark - Legal49.9
Putnam 2025 (MathArena Accuracy)77.5
SimpleBench61.6
SpeciEval96.1
SpeciEval — Belief in Animal Sentience6.8
SpeciEval — Land Animal 4Ns4.3
SpeciEval — Sea Animal 4Ns4.3
SpeciEval — Speciesism1.4
VISTA52.4
VoxelBench Text Rating1233
WeirdML (v2)60.4
Loading Atlas data…