Atlas

Models

← All models

GPT-4.5

OpenAI · 2025-02-27

GPT-4.5 is OpenAI's large non-reasoning model, released as a research preview, offering broader world knowledge, improved instruction-following, and more natural conversation than GPT-4o. It was OpenAI's largest and most knowledge-rich GPT model of its time, oriented toward chat and writing rather than chain-of-thought reasoning.

Benchmark scores

BenchmarkScore
A-Fantasia37.7
A-Fantasia Backwards Spelling32.0
A-Fantasia Chess3.0
A-Fantasia Cube Rotation78.0
Agentic Tool Use (Chat)56.3
Agentic Tool Use (Enterprise)63.8
AI2D (OpenVLM)87.2
AidanBench3004
Aider Polyglot44.9
ARC-AGI-110.3
ARC-AGI-20.8
Artificial Analysis Intelligence Index13.6
BrowseComp0.9
Capability136.2
CharXiv Reasoning55.4
CharXiv-D90.0
Confabulations Leaderboard Confabulation Rate11.9
Confabulations Leaderboard Non-Response Rate15.4
Confabulations Leaderboard Weighted Score13.6
Creative Writing v281.7
Creative Writing v3 Elo1227
Creative Writing v3 Rubric Score15.2
Cybench17.5
Elimination Game TrueSkill Mu5.1
EnigmaEval3.2
Epoch Capabilities Index (ECI)137
EQ-Bench 31053
EQ-Bench 3 (Rubric Score)61.3
EQ-Bench 3: Analytic7.2
EQ-Bench 3: Assertive3.9
Loading Atlas data…