OpenAI launched GPT-5.5 on Thursday, pushing its newest frontier mannequin to lots of of thousands and thousands of ChatGPT and Codex customers barely six weeks after the debut of GPT-5.4 — a cadence of upgrades that’s beginning to really feel much less like product launches and extra like rolling infrastructure deployment.
The San Francisco firm introduced the new model to Plus, Professional, Enterprise and Enterprise subscribers in ChatGPT and Codex on 23 April, with GPT-5.5 Professional rolling out concurrently to higher-tier accounts. OpenAI mentioned the API model would observe “very quickly” at US$5 per million enter tokens and US$30 per million output tokens — roughly double the worth of GPT-5.4 — with a 1-million-token context window. GPT-5.5 Professional shall be priced at US$30 and US$180 per million enter and output tokens respectively. In Codex, the mannequin operates with a 400,000-token context window, whereas a brand new Quick mode generates tokens 1.5x faster at 2.5x the price.
Greg Brockman, OpenAI’s co-founder and president, framed the discharge as the subsequent incremental step towards what he and chief government Sam Altman have described as a “superapp” — a multipurpose agent combining ChatGPT, Codex and different instruments right into a single floor. “This mannequin is an actual step ahead in direction of the form of computing that we count on sooner or later — however it’s one step, and we count on to see many sooner or later,” Brockman told reporters on a press call. “It’s a quicker, sharper thinker for fewer tokens in comparison with one thing like 5.4.”
CEO Sam Altman mentioned on X, “We imagine in iterative deployment; though GPT-5.5 is already a sensible mannequin, we count on speedy enhancements. Iterative deployment is an enormous a part of our security technique; we imagine the world shall be finest outfitted to win on the workforce sport of AI resilience this fashion.”
“We love you and we wish you to win”, mentioned Sam Altman. Supply: X
Agentic capabilities take centre stage
The headline positive aspects focus on longer, multi-step work: planning, software use, browser management, doc and spreadsheet creation, and code debugging. OpenAI claims GPT-5.5 tops the Synthetic Evaluation Coding Agent Index at “half the price” of competing frontier coding fashions, and the Codex model has been meaningfully upgraded — it might now work together with net apps, click on by way of pages, take a look at flows, seize screenshots and iterate on what it sees till a job is full. Mixed with final week’s overhaul of Codex for Mac, which added computer-use options that transcend pure agentic coding, OpenAI is quietly assembling the items of a general-purpose desktop agent.
The brand new mannequin benchmarks are spectacular, Supply: OpenAI
Brockman pushed that framing additional in his remarks to reporters. “What is absolutely particular about this mannequin is how rather more it might do with much less steering,” he mentioned throughout a briefing with journalists. “It may take a look at an unclear drawback and work out simply what must occur subsequent. It actually, to me, feels prefer it’s setting the muse for the way we’re going to make use of computer systems, how we’re going to do pc work going ahead.”
OpenAI additionally confirmed that GPT-5.5 is optimised for what it calls “actual work” — analysing information, writing and debugging code, working software program, researching on-line and creating paperwork and spreadsheets. ChatGPT subscribers get entry to 2 variants: GPT-5.5 Considering, which OpenAI says “unlocks quicker assist for tougher issues, with smarter and extra concise solutions,” and GPT-5.5 Professional, which early testers reportedly described as “a step up in each the issue and high quality of labor ChatGPT can tackle, with latency enhancements that make it rather more sensible for demanding duties.”
Industrial urgency underpins the tempo
OpenAI used the launch to reveal recent utilization figures that appeared engineered to rebut a circulating narrative that Anthropic has been profitable the enterprise struggle. The corporate now has greater than 900 million weekly energetic ChatGPT customers and over 50 million subscribers, alongside Four million energetic Codex customers and 9 million paying enterprise customers. The six-week turnaround since GPT-5.4 — launched on 5 March — underscores how far frontier AI has moved from annual product cycles towards one thing nearer to steady deployment.
That cadence raises the stakes for any enterprise constructing on OpenAI’s stack. A doubling of API costs in opposition to GPT-5.Four is materials, although OpenAI argues token effectivity offsets the headline determine. “Whereas GPT-5.5 is priced larger than GPT-5.4,” the corporate mentioned in its launch submit, “it’s each extra clever and rather more token environment friendly.” For builders engaged on agentic workflows the place a single job could contain dozens of software calls and self-correction loops, that effectivity declare shall be examined shortly.
Security posture and the Mythos shadow
OpenAI categorised each GPT-5.5’s organic and chemical capabilities and its cybersecurity capabilities as “Excessive” below its inner Preparedness Framework — a step up from GPT-5.Four however wanting the “Important” threshold that will set off extra restrictive deployment. The corporate mentioned it labored with practically 200 trusted early-access companions forward of launch, added focused testing for superior cyber and biology dangers, and is treating the API rollout extra cautiously than the patron launch as a result of “API deployments require completely different safeguards.”
The security framing lands immediately on territory not too long ago contested by Anthropic, whose Mythos mannequin drew scrutiny for its defensive cybersecurity capabilities earlier than being partially restricted following experiences of unauthorised entry — an issue BNC examined earlier this month. Requested on the press name whether or not GPT-5.5 would match Mythos’s defensive posture, Mia Glaese, a member of OpenAI’s technical workers, mentioned the corporate had “a powerful and lengthy standing technique for our strategy to cyber, and we’ve refined a sturdy strategy to rolling out fashions safely.”
OpenAI can also be opening a trusted-access programme at chatgpt.com/cyber for verified defensive safety work and mentioned it’s working with authorities companions to discover how superior AI can assist the safety of essential infrastructure — from taxpayer information techniques to native energy grids and water provides. It’s a pitch aimed squarely on the US coverage viewers that has spent the previous quarter watching AI governance transfer up the White Home agenda.
What comes subsequent
For the lots of of thousands and thousands of ChatGPT customers who will discover GPT-5.5 with out serious about Preparedness Frameworks or context home windows, the sensible improve is a mannequin that follows directions extra reliably on advanced duties and requires much less hand-holding to provide a completed artefact. For builders, it’s a sharper, extra token-efficient agent at a materially larger sticker value — with the API nonetheless to land.
The bigger sign is considered one of tempo. OpenAI is transport quicker than at any level in its historical past, Anthropic is pushing into adjoining safety-critical domains, and the window between mannequin generations is collapsing from years to weeks. GPT-5.5 is, as Brockman insisted, only one step. The query is what number of extra steps arrive earlier than the superapp imaginative and prescient stops being a speaking level and begins being how most individuals really use their computer systems.
Jason Jones Jason Jones Read More
