openai/gpt-4o-mini-2024-07-18 GPT 4o Mini

gpt-4o-mini currently points to this version.

Released Date: 7/18/2024

Avg. Accuracy:

68.6%

Latency:

3.96s

Performance by Benchmark

Benchmarks

Accuracy

Rankings

LegalBench

76.2%

( 17 / 43 )

CorpFin

55.0%

( 9 / 19 )

CaseLaw

70.8%

( 27 / 39 )

ContractLaw

72.4%

( 7 / 47 )

TaxEval

64.9%

( 16 / 25 )

MedQA

72.4%

( 17 / 24 )

Academic Benchmarks
Proprietary Benchmarks (contact us to get access)

Cost Analysis

Input Cost

$0.15 / M Tokens

Output Cost

$0.60 / M Tokens

Input Cost (per char)

$0.03 / M chars

Output Cost (per char)

$0.14 / M chars

Overview

GPT-4o Mini represents OpenAI’s effort to provide a cheaper, more lightweight version of GPT-4. It offers a compelling balance of performance and cost, making it particularly suitable for production deployments where both quality and economics matter.

Key Specifications

  • Context Window: 128,000 tokens
  • Output Limit: 16,384 tokens
  • Training Cutoff: October 2023
  • Pricing:
    • Input: $0.15 per million tokens
    • Output: $0.60 per million tokens

Performance Highlights

  • Cost Efficiency: Significantly cheaper than GPT-4 while maintaining strong performance
  • Legal Tasks: Shows strong performance on legal reasoning tasks
  • Consistency: Reliable performance across various benchmark categories

Benchmark Results

The model demonstrates competitive performance across our benchmarks:

  • group performance ranking results:

Use Case Recommendations

Best suited for:

  • High-volume production deployments
  • Cost-sensitive applications
  • Tasks requiring balance of performance and efficiency
  • Legal document analysis at scale

Limitations

  • Lower performance ceiling compared to GPT-4o
  • May struggle with highly complex legal and financial tasks

Comparison with Other Models

  • More capable than GPT-3.5 Turbo
  • More cost-effective than GPT-4
  • Competitive with Claude 3.5 Haiku in terms of performance/cost ratio
Join our mailing list to receive benchmark updates on

Stay up to date as new benchmarks and models are released.