GPT-5.5 Is ‘Our Smartest Model Yet,’ Says Company With History of Saying That

OpenAI today released GPT-5.5, what it says is its “smartest and most intuitive to use model yet, and the next step toward a new way of getting work done on a computer.”

The company says the model understands what users are trying to do faster, can carry more of the workload itself, and excels at writing and debugging code, researching online, analyzing data, creating documents, operating software, and moving across tools until a task is finished.

The company published performance numbers from Terminal-Bench 2.0, which tests complex command-line workflows requiring planning, iteration, and tool coordination.

OpenAI said GPT-5.5 outperformed its predecessor on every major coding and agent benchmark the company tested, while using fewer tokens and running at the same speed as the older model. On one third-party coding index, it matched leading rivals at about half the cost.

Keeping a larger model that fast required rebuilding inference as a single system rather than a patchwork of tweaks, the company said. GPT-5.5 was designed, trained and served on NVIDIA’s latest hardware, and OpenAI credited its own Codex tool and GPT-5.5 itself with helping hit the efficiency targets.

Early testers told the company the model seems to grasp how a codebase fits together — why something is failing, where the fix belongs and what else the change will touch.

Dan Shipper, Founder and CEO of Every, called GPT-5.5 “the first coding model I’ve used that has serious conceptual clarity.” After launching an app, he spent days debugging a post-launch issue before bringing in one of his best engineers to rewrite part of the system. To test GPT-5.5, he effectively rewound the clock: could the model look at the broken state and produce the same kind of rewrite the engineer eventually decided on? GPT-5.4 could not. GPT-5.5 could.

Pietro Schirano, CEO of MagicPath, saw a similar step change when GPT-5.5 merged a branch with hundreds of frontend and refactor changes into a main branch that had also changed substantially — resolving the work in one shot in about 20 minutes.

One engineer at NVIDIA with early access went as far as to say: “Losing access to GPT-5.5 feels like I’ve had a limb amputated.”

OpenAI is already running the model internally at scale. More than 85% of the company uses Codex every week across functions including software engineering, finance, communications, marketing, data science, and product management. The finance team used GPT-5.5 in Codex to review 24,771 K-1 tax forms totaling 71,637 pages, accelerating the task by two weeks compared to the prior year.

The model also shows gains on scientific and technical research workflows. On GeneBench, a new eval focusing on multi-stage scientific data analysis in genetics and quantitative biology, GPT-5.5 outperforms GPT-5.4 on problems that often correspond to multi-day projects for scientific experts. On BixBench, a benchmark designed around real-world bioinformatics and data analysis, it achieved leading performance among models with published scores.

In a notable example, an internal version of GPT-5.5 with a custom harness helped discover a new proof about Ramsey numbers — one of the central objects in combinatorics — later verified in the Lean proof assistant. The result is a concrete example of GPT-5.5 contributing not just code or explanation, but a novel mathematical argument in a core research area.

OpenAI says GPT-5.5 was released with its strongest safeguards to date, including tighter controls around cybersecurity workflows and repeated misuse patterns. The model was evaluated across the company’s full safety and preparedness frameworks, with input from nearly 200 trusted early-access partners before launch.

GPT-5.5 is available today in ChatGPT and Codex for Plus, Pro, Business, and Enterprise users. GPT-5.5 Pro is rolling out to Pro, Business, and Enterprise tiers. API access, which requires different safeguards, is coming “very soon,” OpenAI said.