GPT-4 (source unspecified)

Name: GPT-4 (source unspecified)
Author: OpenAI

OpenAI

This benchmark entry retains GPT-4 observations whose source does not identify a dated checkpoint. OpenAI changed the gpt-4 API alias from the March 14 to the June 13, 2023 snapshot; those exact artifacts have separate model rows, and this source-unspecified entry does not represent a third released checkpoint.

Benchmark scores

Benchmark	Score
AA-Omniscience Index	-33.8
Agentic Tool Use (Chat)	37.9
Agentic Tool Use (Enterprise)	51.4
AlpacaEval 2.0 (length-controlled win rate)	38.1
AlpacaEval 2.0 (raw win rate)	23.6
Artificial Analysis Coding Index	13.1
Artificial Analysis Omniscience Accuracy	21.0
Artificial Analysis Omniscience Hallucination Rate	69.4
Artificial Analysis Progress Index v1	140
BBH (BIG-Bench Hard)	83.1
Epoch Capabilities Index (ECI)	126
GDPval-AA v2	189
GDPval-AA v2 (normalized)	0.0
GDPval-AA v2.1	-12.4
GPQA Diamond (Artificial Analysis avg@5)	34.9
IFBench (Artificial Analysis avg@5)	33.2
MATH	52.9
MATH-500 (Artificial Analysis source)	56.8
MGSM	74.5
MMLU (uncertainty-routed CoT@32)	87.3
MMLU Pro	56.2
ObviousBench	72.9
Ruby LLM Benchmarks — Program Fixer	63.8
Ruby LLM Program Fixer — Calendar	81.5
Ruby LLM Program Fixer — Parking Garage	39.8
Ruby LLM Program Fixer — School Library	55.8
Ruby LLM Program Fixer — Vending Machine	77.9