Key takeaways

• GPT-5 (minimal) is a proprietary, Transformer-based model by OpenAI, designed for complex tasks requiring advanced reasoning, instruction following, and high accuracy, optimized for code quality and user experience in high-stakes use cases.
• The model utilizes advanced prompt understanding and test-time routing features, operating with a text+image-to-text modality.
• GPT-5 (minimal) excels in coding, writing, and health-related tasks, demonstrating reductions in hallucination and sycophancy. However, it struggles with incorrect numerical computation, formatting violations, and specifically performed poorly on the 'Terminal-Bench' dataset.
• The model achieved its best performance with 96.6% accuracy on the 'MATH-500' dataset, indicating strong mathematical reasoning abilities due to tasks requiring precise calculations and logical steps.
• Its worst performance was on the 'Terminal-Bench' dataset, with only 26.25% accuracy, highlighting challenges with numerical computation and formatting, suggesting a need for supervised fine-tuning in these areas.

Technical specifications

Context length

400k tokens

Max output

—

Architecture

Transformer

Modalities

text+image->text

Parameters

~8000B

Open weights

No

Licence

Proprietary

Regions

USA

Vendor

OpenAI

Provider model key

openai/gpt-5-minimal