GPT-4: Complete Benchmark Performance Guide

openai 4 2023-03-14 Proprietary

In-depth analysis of GPT-4 performance on MMLU, HumanEval, GSM8K, MATH, BBH benchmarks, including pricing and use cases.

Key Specifications

Vendoropenai
Version4
Release Date2023-03-14
Context Window8192 tokens
Input Modalitiestext
Output Modalitiestext
LicenseProprietary
Documentationhttps://platform.openai.com/docs/models

Benchmark Performance

BenchmarkScoreUnitEvaluated AtNotesSource
MMLU85.8%2023-03-145-shotview
HUMANEVAL77.6pass@12023-03-14—view
GSM8K91.7%2023-03-140-shot CoTview
MATH43.9%2023-03-140-shot CoTview
BBH80.9%2023-03-143-shot CoTview
GPQA39.2%2023-03-140-shotview
IFEVAL74.6%2023-03-14prompt_strictview
ARC94.1%2023-03-14challengeview
MUSR54.6%2023-03-140-shotview
WINOGRANDE80.3%2023-03-140-shotview

Pricing

TierPriceCurrency
Input$30 / MtokUSD
Output$60 / MtokUSD
Cache Read$0 / MtokUSD
Cache Write$0 / MtokUSD

Source: https://openai.com/api/pricing/ · as of 2023-03-14

Compliance

  • Data Residency: US
  • SOC2: ✓
  • HIPAA: ✓
  • GDPR: ✓
  • ISO 27001: ✓

GPT-4

Model Overview

OpenAI GPT-4 大型语言模型, 8K 上下文, 在推理/创意/协作任务上显著超越 GPT-3.5, 首次引入人类水平表现。

Core Specifications

VendorVersionRelease DateContext WindowInput ModalitiesOutput ModalitiesLicense
Openai42023-03-148KtexttextProprietary

Benchmark Performance

BenchmarkScoreUnitNotes
MMLU (Massive Multitask Language Understanding)85.8%5-shot
HumanEval77.6pass@1—
GSM8K (Grade School Math 8K)91.7%0-shot CoT
MATH43.9%0-shot CoT
BBH (BIG-Bench Hard)80.9%3-shot CoT
GPQA39.2%0-shot
IFEval74.6%prompt_strict
ARC94.1%challenge
MUSR54.6%0-shot
WinoGrande80.3%0-shot

Pricing

InputOutputCache ReadCache Write
————

per million tokens

Strengths

  • MMLU score 85.8, strong knowledge reasoning.
  • GSM8K 91.7, robust math reasoning.

Weaknesses

  • Proprietary, not self-hostable.
  • Context window 8K is limited.

Use Cases

  • Code generation and debugging
  • Agent workflows and tool use

References