Model

GPT-4

CLOSED

OpenAI · GPT-4 family · released Mar 14, 2023

First GPT-4; multimodal text+image input, 8K default context (32K gpt-4-32k variant, since deprecated); large reasoning jump over GPT-3.5.

ReasoningCodingVisionFunction callingTool useAgentic

1037.4

Elo · rank #207

Parameters

Undisclosed

Active params

Undisclosed

Context

8K tokens

Architecture

Transformer; widely reported (unconfirmed by OpenAI) as an ~8x220B mixture-of-experts

License

Proprietary

Languages

—

API price (in/out)

$30 / $60

Modalities

text · vision

Benchmark results

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

ARC-ChallengeReasoning96.3%#4

best: Llama 3.1 405B · 96.9%

CodeforcesCoding392#44

best: DeepSeek-V4-Pro (Think Max) · 3206

DROPReasoning80.9#23

best: Hunyuan-T1 · 93.1

GSM8KMath92.0%#31

best: Llama 3.1 405B · 96.8%

HellaSwagReasoning95.3%#2

best: Claude 3 Opus · 95.4%

APIBench (HuggingFace)Agents19.8%#2

best: Gorilla 7B · 71.7%

HumanEvalCoding67.0%#89

best: Claude Opus 4.5 · 99.4%

LogicKorHuman preference8.67#10

best: GPT-4o · 9.33

MMLUKnowledge86.4%#43

best: OpenAI o3 · 92.9%

OpenAssistant BenchHuman preference1294#1

best: this model · 1294

APIBench (TensorHub)Agents18.2%#3

best: Gorilla 7B · 83.8%

APIBench (TorchHub)Agents38.7%#3

best: Gorilla 7B · 59.1%

Vicuna BenchHuman preference1176#1

best: this model · 1176

WinoGrandeReasoning87.5%#6

best: PaLM 2 · 90.9%

Run it locally

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

Input / M tok

$30

Output / M tok

$60

GPT-4 family

Elo progression across releases

GPT-4Mar 20231037.4 GPT-4 TurboNov 20231142.7

Resources

DOCSdevelopers.openai.com/api/docs/models/gpt-4 APIAvailable via OpenAI API

API price $30/$60 · each benchmark row carries its own source badge (see methodology)