OpenAI introduces GPT-5.5 for ChatGPT and Codex Apfelpatient

This article may contain affiliate links. Learn more on our Partner Program page.

Make Apfelpatient a preferred source One click – and you'll see us more often on Google

OpenAI has released a new version of its language model, GPT-5.5, enhancing both ChatGPT and Codex. The company promises a “new class of intelligence for real work” – clearly focusing on agent-based tasks, multi-stage workflows, and professional applications.

OpenAI's rapid pace continues. Seven weeks after releasing GPT-5.4 Thinking in early March, the company is now following up with GPT-5.5. This is in keeping with OpenAI's usual speed: since GPT-5 in August 2025, followed by GPT-5.1 in November, GPT-5.2 as a response to Google's Gemini 3, and GPT-5.3 Instant in March, this is already the fifth major update in eight months.

With GPT-5.5, OpenAI shifts the focus further away from mere conversation quality and towards genuine tool utilization. According to OpenAI, the new model "understands faster what you're trying to do" and can handle more tasks automatically. This sounds abstract, but has concrete consequences: Anyone who uses ChatGPT or Codex professionally for complex tasks will gain a tool with GPT-5.5 that independently plans multiple work steps, uses tools, and checks its own work.

The four key improvements

OpenAI highlights four areas where GPT-5.5 is said to have improved significantly:

Agentic coding: The model can independently plan, implement, and test code projects.
Computer usage: GPT-5.5 can operate applications, navigate websites, and evaluate screenshots.
Knowledge work: Analyzing complex documents, recognizing connections, and arguing in multiple stages.
Early scientific research: formulating hypotheses, analyzing data sets, and testing results

What's particularly exciting about Codex is its computer-use aspect. OpenAI has extended browser usage so that Codex can interact with web apps, run test flows, and click through pages – and then iteratively improve until the task is complete. This goes far beyond traditional coding and directly aims at the vision of an "AI agent" that performs real work in the browser.

Thinking and Pro: Two variants

As with previous versions, GPT-5.5 comes in two editions. According to OpenAI, GPT-5.5 Thinking is designed to provide faster help with difficult problems, offering more precise and concise answers. GPT-5.5 Pro is intended for even more demanding tasks: higher quality, greater depth in complex questions. According to OpenAI, early testers particularly praised the latency improvements, which make Pro more practical for demanding tasks.

Availability is staggered: GPT-5.5 is being rolled out immediately to Plus, Pro, Business, and Enterprise users in ChatGPT and Codex. GPT-5.5 Pro is only available to Pro, Business, and Enterprise customers. Codex also introduces a Fast Mode that generates tokens 1.5 times faster – but at 2.5 times the cost.

Prices are rising – but there's a catch

GPT-5.5 is more expensive than GPT-5.4. OpenAI argues, however, that the model is both more intelligent and significantly more token-efficient. According to OpenAI, GPT-5.5 should deliver better results with fewer tokens for Codex users than GPT-5.4 – partially offsetting the higher Pro costs. Nevertheless, the update will mean higher costs for many developers if they want to use the new capabilities.

For end users in ChatGPT, access via existing subscriptions (Plus, Pro, Business, Enterprise) remains unchanged – the monthly fee remains the same.

Safety focus: Two weeks of red teaming

One sentence in OpenAI's announcement deserves special attention: GPT-5.5 has been shipped with the strictest protection mechanisms to date. Prior to its release, the model was tested by internal and external red teams, particularly with regard to cybersecurity and biological capabilities. Nearly 200 trusted early access partners contributed feedback from real-world use cases.

This is noteworthy because OpenAI only recently unveiled GPT-5.4-Cyber for cyber defense – a specialized model specifically designed for security applications. With GPT-5.5, OpenAI is also tightening the security measures for its main model. This isn't just PR: More powerful AI models can also be more dangerous if misused. The more agent-like a model becomes, the more crucial its security measures become.

A week full of announcements

GPT-5.5 marks the end of an intense week for OpenAI. Just two days ago, the company released ChatGPT Images 2 for significantly improved image generation – a model with thinking capabilities that makes image results noticeably more precise. Last week also saw a major Codex update with computer-use features.

Together, this reveals a clear pattern: OpenAI is consistently moving from purely chat- and image-based models to agent-based systems that can independently perform tasks. The vision behind this is an AI system that functions like a digital employee – with access to browsers, files, tools, and web services.

Availability and outlook

GPT-5.5 is now available for Plus, Pro, Business, and Enterprise users in ChatGPT and Codex. In Codex, the model is available with a 400K context window. OpenAI has announced that GPT-5.5 and GPT-5.5 Pro will also be available in the API "very soon." Developers building their own products based on GPT will then also gain access to the new capabilities – at a correspondingly higher cost than with GPT-5.4.

The pace at which OpenAI is developing its models shows no signs of slowing down. Anyone actively following AI development should expect the next version to be released by summer at the latest. (Image: Shutterstock / TY Lim)