Models / OpenAI

GPT-4 (2023)

GPT-4 · Released 2023-03-14

Original GPT-4 technical report baseline. MMLU 86.4%, HumanEval 67.0% pass@1. Anchor for pre-2024 frontier comparisons.

Historical anchor only — not deployable on current OpenAI routes.

Composite67

Context8K

Input / 1M$30

Output / 1M$60

Knowledge cutoff2021-09-01

Statuslightweight

Compare with:

Knowledge: 86
Math: 53
Coding: 67

69average across 3 tested families

Benchmark placements

Per-benchmark positions

Where GPT-4 (2023) places on each public benchmark source that publishes a row. Rank counts every model with a latest row in the same test — not a universal quality score.

Benchmark	Family	Rank	Score	Source	Date
HumanEval	Coding	#15/ 15	67	GPT-4 Technical Report	2023-03-15
MMLU	Knowledge	#17/ 22	86	GPT-4 Technical Report	2023-03-15
MATH	Math	#20/ 20	53	GPT-4 Technical Report	2023-03-15

Family context

Strongest rows in each family

The three highest-scoring carded models in each capability family. Where GPT-4 (2023) shows up, it's highlighted.

KnowledgeGPT-4 (2023) · 86

Reasoning

MathGPT-4 (2023) · 53

CodingGPT-4 (2023) · 67

Agentic

1
GPT 5.5
100
2
Claude Opus 4.8
100
3
Claude Fable 5
100

Multimodal

Long context

1
Kimi K3
75
2
GPT-5.1
75
3
GPT-5.6 Sol
74

Tool use

Safety

Human preference

Family coverage

Best score per family

A high composite that hides a weak family is a trap. These bars surface the families where this model hasn't been publicly tested, and where it leads.

Knowledge86

Math53

Coding67

Price vs. performance