OpenAI has released GPT-5.5, presenting it as its newest and most capable AI model for real-world work across coding, research, data analysis, and office tasks. The GPT-5.5 launch centers on a bigger shift toward “agentic” AI, meaning a system that can handle complex assignments with less hand-holding by planning steps, using tools, checking its own output, and continuing through unclear or messy instructions.
The company says GPT-5.5 is rolling out in ChatGPT and Codex for Plus, Pro, Business, and Enterprise users, while GPT-5.5 Pro is reserved for Pro, Business, and Enterprise tiers. OpenAI has also set API pricing at $5 per 1 million input tokens and $30 per 1 million output tokens, though reports differ slightly on timing, with some saying wider API access is expected soon and others saying it will be released separately later.
Built for longer tasks
OpenAI is positioning GPT-5.5 as a model that can move beyond one-shot answers and manage work that unfolds over multiple steps. Reports say it can write and debug code, analyze datasets, create reports, spreadsheets, and presentations, and work across software environments more smoothly than earlier versions.
That broader scope is a major part of the company’s pitch. OpenAI says the model is designed for areas such as agentic coding, computer use, general office work, and early scientific research, all of which require reasoning over longer contexts and carrying tasks forward over time. The company has also said internal teams are already using it across finance, communications, and product management.
Benchmarks and efficiency
Several published benchmark results are being used to highlight the upgrade. Reported scores include 82.7% on Terminal-Bench 2.0, 84.9% on GDPval for knowledge work, and 78.7% on OSWorld for computer task execution. One report also said GPT-5.5 Pro scored 39.6% on FrontierMath Tier 4, compared with 22.9% for Anthropic’s Claude Opus 4.7.
OpenAI is also stressing efficiency, not just raw capability. Company executives described GPT-5.5 as a faster, sharper thinker that uses fewer tokens than GPT-5.4, while maintaining similar real-world latency. That means the model is being marketed as both more capable and more practical for business use, especially for teams that want better output without slower response times.
Some early-access users cited by reports said the model helped them review vibe-coded work, examine thousands of extra documents, and save up to 10 hours of work a week. Other coverage says GPT-5.5 handles large codebases more effectively, shows stronger persistence on longer engineering tasks, and is better at spotting issues before they become bigger problems.
Enterprise push
The release also fits into OpenAI’s bigger product strategy. Greg Brockman said GPT-5.5 moves the company closer to its planned “super app,” a unified service that would combine ChatGPT, Codex, and an AI browser for enterprise customers. That framing suggests OpenAI is not just launching another model, but building toward a broader software platform.
The timing matters as well. The launch came about a week after Anthropic introduced its own new model, and several reports framed GPT-5.5 as part of a faster-moving AI race focused on enterprise adoption. OpenAI’s message is that businesses and professionals want systems that can do more than answer questions; they want tools that can carry out work from start to finish.
Safety measures
OpenAI says GPT-5.5 includes its strongest safeguards so far, with testing across cybersecurity and biological risk frameworks and tighter controls for sensitive use cases. At the same time, the company is expanding access for verified users in defensive security work, showing that it is trying to balance stronger protections with more advanced uses.
Alongside the launch, OpenAI opened a restricted GPT-5.5 Bio Bug Bounty focused on finding a single universal jailbreak that can answer five bio-safety questions from a clean chat in Codex Desktop without triggering moderation. The top reward is $25,000, with smaller discretionary awards for partial findings. Applications opened on April 23 and close on June 22, while the testing window runs from April 28 to July 27.
That program adds another layer to the rollout story. GPT-5.5 is being introduced as a model built for longer, more autonomous workflows, but OpenAI is also signaling that safety testing remains a central part of how it wants these systems deployed. Together, the product launch, benchmark claims, enterprise positioning, and bug bounty show how OpenAI is trying to push more capable AI into everyday work while tightening its safeguards around higher-risk use cases.
