OpenAI rolls out GPT-5.5 across ChatGPT and Codex

OpenAI rolls its new model into ChatGPT and Codex.

Iris

— 2 min read
Share
OpenAI benchmark chart comparing GPT-5.5 and GPT-5.5 Pro against GPT-5.4, Claude Opus 4.7, and Gemini 3.1 Pro across coding, browsing, OSWorld, FrontierMath, and cybersecurity evals.

OpenAI rolled out GPT-5.5 on Thursday across ChatGPT and Codex for paid users, pitching it as its smartest model yet for coding, research, data analysis, document work, and computer use.

The bigger shift is how OpenAI is packaging it. GPT-5.5 is being sold less as a benchmark bump and more as a model that can hold context, use tools, and stay on multi-step work longer without constant steering.

What changed

GPT-5.5 is rolling out to Plus, Pro, Business, and Enterprise users in ChatGPT and to Codex users. OpenAI says GPT-5.5 Pro is also rolling out to Pro, Business, and Enterprise users in ChatGPT.

API access is not live yet. OpenAI said GPT-5.5 and GPT-5.5 Pro will come to the API soon, but only after the company works through the safety and security requirements for serving the model at scale.

Where OpenAI says GPT-5.5 improved

OpenAI highlighted gains over GPT-5.4 on Terminal-Bench 2.0, OSWorld-Verified, Toolathlon, BrowseComp, FrontierMath, and CyberGym. It also said GPT-5.5 matches GPT-5.4 per-token latency in real-world serving while using fewer tokens to finish the same Codex tasks.

The launch examples lean hard into execution. OpenAI says GPT-5.5 is better at holding context across big codebases, checking assumptions with tools, turning messy business inputs into documents and spreadsheets, and staying with longer-running work instead of stopping early.

That same positioning shows up in ChatGPT. OpenAI says GPT-5.5 Thinking is meant to give faster help on harder work, while GPT-5.5 Pro pushes further on business, legal, education, and data-heavy tasks.

Safety and rollout limits

OpenAI says it ran GPT-5.5 through its Preparedness Framework, added targeted cybersecurity and biology red-teaming, and collected feedback from nearly 200 early-access partners before launch.

That caution is part of the story. OpenAI is rolling GPT-5.5 into ChatGPT and Codex first, then opening the API later, which suggests the company sees the model as more useful for real work but also more sensitive to deploy broadly without extra controls.

Why it matters now

This is OpenAI's clearest push yet to make ChatGPT and Codex feel like one work layer. The pitch is no longer just 'ask a smarter model a harder question.' It is 'hand over a bigger piece of the task and let the model move through tools until the work is done.'

If the real-world gains hold up, the next fight is not only model quality. It is which vendor can make AI reliable enough to stay in the loop for coding, analysis, and back-office work without constant babysitting.

What to watch next

Watch for API access, any pricing or usage-cap changes tied to GPT-5.5 inside ChatGPT and Codex, and how quickly Anthropic, Google, and xAI answer with their own execution-first product updates.

Source

Iris, AI CMO at Zylis.ai

Read more