OpenAI introduces GPT-5.4: More performance for ChatGPT

AI company OpenAI has announced another major update for ChatGPT. With GPT-5.4 Thinking and an additional Pro variant, the company's most powerful frontier model to date is being significantly enhanced. The goal is to substantially expand the AI's capabilities in key areas.

Just a few days ago, OpenAI released a new version, GPT-5.3 Instant, promising to make ChatGPT feel more natural and reduce embarrassing errors. Now, GPT-5.4 Thinking represents the next step. According to OpenAI, the update brings six key improvements in areas such as programming, document analysis, multimodal tasks, and agent-based workflows.

The development of modern AI models is currently progressing at a rapid pace. Especially in competition with other large AI providers, models are regularly expanded and improved.

OpenAI has also revised its main model several times in recent months. Most recently, this happened in December with GPT-5.2, which was internally released as "Code Red" in response to the growing competition from Google's Gemini systems.

With GPT-5.4 Thinking, OpenAI continues this development. The new model is particularly aimed at applications with complex workflows, large datasets, and automated processes. In addition to functional improvements, the update also brings technical innovations such as a significantly expanded context window and integrated computing functions.

The six most important improvements of OpenAI GPT-5.4

OpenAI identifies a total of six key areas where GPT-5.4 Thinking offers improvements.

Improved programming, document analysis, and tool usage

The new model is more accurate in programming tasks and can analyze large documents more effectively. Its handling of external tools has also been optimized. Furthermore, GPT-5.4 follows instructions more precisely than previous versions.

These improvements ensure that complex tasks can be implemented more reliably, especially when multiple tools or data sources are involved.

Advances in image processing and multimodal tasks

OpenAI has also improved the model's image perception and multimodal capabilities. GPT-5.4 can better interpret information from images and combine it with text data.

This makes it possible to solve tasks where different data types need to be analyzed together.

Improved execution of complex and lengthy tasks

Another focus is on the execution of long-running tasks and multi-stage agent workflows.

The model can coordinate multiple steps within a process and execute complex task chains over extended periods. This capability is particularly important for automated systems that independently organize various work steps.

More efficient token usage and higher overall performance

Efficiency has also been improved. GPT-5.4 utilizes tokens more efficiently, thus offering better end-to-end performance for tool-intensive workloads.

This applies especially to applications where many external systems, APIs, or tools are integrated into the workflow.

Agentic web search and multi-source synthesis

Another focus is on internet research. GPT-5.4 supports agentic web search, where information from multiple sources is collected and combined.

Especially when information is difficult to find, the model can combine different data sources and create a structured answer from them.

Support for document- and spreadsheet-intensive business processes

OpenAI also highlights enterprise applications. GPT-5.4 is specifically optimized for workflows that work with many documents or spreadsheets. Typical use cases include:

Customer service
Data analysis
Financial processes

Here, the model can analyze large amounts of data and deliver structured results.

Context window of up to one million tokens

One of the most important technical innovations of GPT-5.4 is the greatly enlarged context window of up to one million tokens.

This large context window makes it possible to process very large amounts of data in a single query. This includes, for example:

complete codebases
large document collections
extensive agent trajectories

This allows complex analyses to be performed without having to split content across multiple individual requests.

Integrated computer use in the OpenAI main model

GPT-5.4 is also OpenAI's first mainline model with integrated computer use.

This feature allows AI agents to interact directly with software. The model can process tasks in a so-called build-run-verify-fix cycle. This involves first creating a solution, then executing it, subsequently verifying it, and automatically correcting it if necessary.

This approach allows complex processes to be tested and improved automatically.

Training for compression and longer agent trajectories

Another new feature concerns the training of the model. According to OpenAI, GPT-5.4 was specifically trained to support compression.

This allows the model to process longer agent trajectories while preserving important context. Even with long workflows, the relevant core information is retained.

Availability of GPT-5.4 in ChatGPT

OpenAI is currently beginning the phased rollout of GPT-5.4.

The model is integrated into both ChatGPT and Codex. The GPT-5.4 Thinking variant is initially available for the following subscriptions:

Plus
team
Per

At the same time, OpenAI announces that GPT-5.2 Thinking will be discontinued within three months.

GPT-5.4: Another development step for OpenAI

With GPT-5.4 Thinking, OpenAI significantly expands the capabilities of ChatGPT. The update not only improves existing features but also introduces new technical possibilities.

The large context window, integrated computer usage, and optimized agent workflows, in particular, demonstrate that OpenAI is increasingly aligning its AI systems for complex tasks and automated processes.

The phased rollout of ChatGPT and Codex marks another step in the further development of the OpenAI platform and its AI models. (Image: Shutterstock / Primakov)