OpenAI officially introduces GPT-5.4 mini and nano

OpenAI is expanding its model portfolio once again, introducing two new variants, GPT-5.4 mini and GPT-5.4 nano, specifically designed for efficiency, speed, and scalability. According to OpenAI, both models are among the company's most powerful small AI systems to date. While GPT-5.4 mini is available immediately for ChatGPT users, GPT-5.4 nano is primarily aimed at developers and API applications.

The development of smaller, high-performance models is becoming increasingly important. Not every application requires maximum model size; often, response time, cost, and stable performance under heavy load are crucial. With its new versions, OpenAI clearly aims to meet precisely these requirements.

This release is part of a series of recent updates with which OpenAI is continuously expanding and optimizing its ecosystem. This focuses not only on pure performance improvements, but also on practical applications in real-world work environments.

GPT-5.4 mini: Significantly faster and more powerful

GPT-5.4 mini represents a clear further development of GPT-5 mini and brings noticeable improvements in several key areas:

Programming: more precise and reliable code creation
Logical thinking: better results in complex tasks
Multimodal understanding: improved processing of different data types
Tool usage: more efficient integration of external tools

A significant improvement lies in the speed. GPT-5.4 mini operates more than twice as fast as its predecessor. Despite its smaller size, it nearly matches the performance of the larger GPT-5.4 model in several benchmarks, including:

SWE-Bench Pro
OSWorld-Verified

This positions GPT-5.4 mini as a high-performance solution for high-workload applications where both speed and quality are crucial.

Availability and use in ChatGPT

GPT-5.4 mini is widely available and is integrated into several areas:

In ChatGPT, it is available to Free and Go users via the "Thinking" function in the Plus menu.
For other users, it serves as a fallback model in case of rate limitations of GPT-5.4 Thinking.
Additionally, it is usable in Codex and via the OpenAI API.

This broad integration ensures that both end users and developers can directly access the model.

GPT-5.4 nano: Focus on speed and cost

In addition to the mini version, OpenAI is also introducing the smallest variant of the new model series, the GPT-5.4 nano. This model is specifically designed for scenarios where low cost and high speed are paramount.

GPT-5.4 nano represents a significant improvement over GPT-5 nano and is particularly suitable for:

Classification tasks
Data extraction
Ranking processes
Development of subagents for simple support tasks

Especially in agent-based workflows, GPT-5.4 nano specifically takes over smaller tasks, thereby relieving larger, computationally intensive models.

The model is available via the OpenAI API and is primarily aimed at developers and businesses.

Classification within current OpenAI developments

The introduction of GPT-5.4 mini and nano is part of a larger series of innovations at OpenAI:

GPT-5.4 Thinking was released at the beginning of the month, offering six key improvements.
Also released in March was GPT-5.3 Instant, which is designed to generate fewer inappropriate or "embarrassing" responses.
OpenAI released the Codex developer app for Mac back in February.

These developments show that OpenAI is not only working on pure model performance, but also on specialization, reliability and practical integration into different working environments.

OpenAI relies on efficient and scalable AI models

With GPT-5.4 mini and GPT-5.4 nano, OpenAI further expands its range of efficient, scalable AI models. Both variants address different requirements: While GPT-5.4 mini serves as a powerful all-rounder for more demanding tasks, GPT-5.4 nano is designed for maximum efficiency in simpler processes.

The new models illustrate the growing importance of compact AI systems. Speed, cost control, and flexible deployment options are the focus of further development. (Image: OpenAI)