OpenAI Ships GPT-5.5 as Agentic AI Race Turns Into a Weekly Sprint

The San Francisco company introduced the new model to Plus, Pro, Business and Enterprise subscribers in ChatGPT and Codex on 23 April, with GPT-5.5 Pro rolling out simultaneously to higher-tier accounts. OpenAI said the API version would follow “very soon” at US$5 per million input tokens and US$30 per million output tokens — roughly double the price of GPT-5.4 — with a 1-million-token context window. GPT-5.5 Pro will be priced at US$30 and US$180 per million input and output tokens respectively. In Codex, the model operates with a 400,000-token context window, while a new Fast mode generates tokens 1.5x quicker at 2.5x the cost.

Greg Brockman, OpenAI’s co-founder and president, framed the release as the next incremental step toward what he and chief executive Sam Altman have described as a “superapp” — a multipurpose agent combining ChatGPT, Codex and other tools into a single surface. “This model is a real step forward towards the kind of computing that we expect in the future — but it is one step, and we expect to see many in the future,” Brockman told reporters on a press call. “It’s a faster, sharper thinker for fewer tokens compared to something like 5.4.”

CEO Sam Altman said on X, “We believe in iterative deployment; although GPT-5.5 is already a smart model, we expect rapid improvements. Iterative deployment is a big part of our safety strategy; we believe the world will be best equipped to win at the team sport of AI resilience this way.”

“We love you and we want you to win”, said Sam Altman. Source: X

Agentic capabilities take centre stage

The headline gains concentrate on longer, multi-step work: planning, tool use, browser control, document and spreadsheet creation, and code debugging. OpenAI claims GPT-5.5 tops the Artificial Analysis Coding Agent Index at “half the cost” of competing frontier coding models, and the Codex version has been meaningfully upgraded — it can now interact with web apps, click through pages, test flows, capture screenshots and iterate on what it sees until a task is complete. Combined with last week’s overhaul of Codex for Mac, which added computer-use features that go beyond pure agentic coding, OpenAI is quietly assembling the pieces of a general-purpose desktop agent.

The new model benchmarks are impressive, Source: OpenAI

Brockman pushed that framing further in his remarks to reporters. “What is really special about this model is how much more it can do with less guidance,” he said during a briefing with journalists. “It can look at an unclear problem and figure out just what needs to happen next. It really, to me, feels like it’s setting the foundation for how we’re going to use computers, how we’re going to do computer work going forward.”

OpenAI also confirmed that GPT-5.5 is optimised for what it calls “real work” — analysing data, writing and debugging code, operating software, researching online and creating documents and spreadsheets. ChatGPT subscribers get access to two variants: GPT-5.5 Thinking, which OpenAI says “unlocks faster help for harder problems, with smarter and more concise answers,” and GPT-5.5 Pro, which early testers reportedly described as “a step up in both the difficulty and quality of work ChatGPT can take on, with latency improvements that make it much more practical for demanding tasks.”

Commercial urgency underpins the pace

OpenAI used the launch to disclose fresh usage figures that appeared engineered to rebut a circulating narrative that Anthropic has been winning the enterprise fight. The company now has more than 900 million weekly active ChatGPT users and over 50 million subscribers, alongside 4 million active Codex users and 9 million paying business users. The six-week turnaround since GPT-5.4 — released on 5 March — underscores how far frontier AI has moved from annual product cycles toward something closer to continuous deployment.

That cadence raises the stakes for any business building on OpenAI’s stack. A doubling of API prices against GPT-5.4 is material, though OpenAI argues token efficiency offsets the headline figure. “While GPT-5.5 is priced higher than GPT-5.4,” the company said in its launch post, “it is both more intelligent and much more token efficient.” For developers working on agentic workflows where a single task may involve dozens of tool calls and self-correction loops, that efficiency claim will be tested quickly.

Safety posture and the Mythos shadow

OpenAI classified both GPT-5.5’s biological and chemical capabilities and its cybersecurity capabilities as “High” under its internal Preparedness Framework — a step up from GPT-5.4 but short of the “Critical” threshold that would trigger more restrictive deployment. The company said it worked with nearly 200 trusted early-access partners ahead of release, added targeted testing for advanced cyber and biology risks, and is treating the API rollout more cautiously than the consumer launch because “API deployments require different safeguards.”

The safety framing lands directly on territory recently contested by Anthropic, whose Mythos model drew scrutiny for its defensive cybersecurity capabilities before being partially restricted following reports of unauthorised access — a controversy BNC examined earlier this month. Asked on the press call whether GPT-5.5 would match Mythos’s defensive posture, Mia Glaese, a member of OpenAI’s technical staff, said the company had “a strong and long standing strategy for our approach to cyber, and we’ve refined a durable approach to rolling out models safely.”

OpenAI is also opening a trusted-access programme at chatgpt.com/cyber for verified defensive security work and said it is working with government partners to explore how advanced AI can support the protection of critical infrastructure — from taxpayer data systems to local power grids and water supplies. It is a pitch aimed squarely at the US policy audience that has spent the past quarter watching AI governance move up the White House agenda.

What comes next

For the hundreds of millions of ChatGPT users who will notice GPT-5.5 without thinking about Preparedness Frameworks or context windows, the practical upgrade is a model that follows instructions more reliably on complex tasks and requires less hand-holding to produce a finished artefact. For developers, it is a sharper, more token-efficient agent at a materially higher sticker price — with the API still to land.

The larger signal is one of tempo. OpenAI is shipping faster than at any point in its history, Anthropic is pushing into adjacent safety-critical domains, and the window between model generations is collapsing from years to weeks. GPT-5.5 is, as Brockman insisted, just one step. The question is how many more steps arrive before the superapp vision stops being a talking point and starts being how most people actually use their computers.