Model

GPT-4.5

CLOSED

OpenAI · GPT-4.5 family · released Feb 27, 2025

'Orion' — OpenAI's largest non-reasoning model; shipped as a costly research preview, removed from API 2025-07-14 and fully sunset from ChatGPT 2026-06-27.

ReasoningCodingVisionFunction callingTool useAgentic

1479.2

Elo · rank #126

Parameters

Undisclosed

Active params

—

Context

128K tokens

Architecture

Dense pretrained transformer (size undisclosed; non-reasoning)

License

Proprietary

Languages

—

API price (in/out)

$75 / $150

Modalities

text · vision

Benchmark results

Bar shows position within the tracked field; marker = field best

Aider PolyglotCoding44.9%#28

best: Claude Opus 4.5 · 89.4%

AIMEMath36.7%#128

best: GPT-5.2 · 100.0%

Arena-Hard v2Human preference50.0%#7

best: Qwen3-Max · 86.1%

Arena EloHuman preference1411#20

best: Claude Fable 5 · 1505

GPQA DiamondReasoning71.4%#117

best: GPT-5.6 · 94.6%

MMLUKnowledge89.6%#13

best: OpenAI o3 · 92.9%

MMMLUKnowledge85.1%#19

best: Gemini 3.1 Pro · 92.6%

MMMUVision74.4%#38

best: Claude Fable 5 · 89.3%

SimpleQAKnowledge62.5%#1

best: this model · 62.5%

SWE-bench VerifiedCoding38.0%#106

best: Claude Fable 5 · 95.0%

Run it locally

Closed weights — available via API only. No local deployment.

Input / M tok

$75

Output / M tok

$150

GPT-4.5 family

Elo progression across releases

GPT-4.5Feb 20251479.2

Resources

DOCSdevelopers.openai.com/api/docs/models/gpt-4.5-preview

API price $75/$150 · each benchmark row carries its own source badge (see methodology)