GPT-5.5 Released: The Smarter AI That Actually Gets Work Done

Core Upgrade: From Answering to Doing

GPT‑5.5’s main improvement is its ability to understand intent. When faced with a chaotic multi‑step task, it automatically plans, calls tools, checks results, and keeps progressing until the task is finished, eliminating the need for users to feed step‑by‑step instructions.

Programming Ability Soars, Beats All Competitors

On the Terminal‑Bench 2.0 complex command‑line workflow benchmark, GPT‑5.5 reaches 82.7% accuracy, far above GPT‑5.4 (75.1%), Claude Opus 4.7 (69.4%) and Gemini 3.1 Pro (68.5%). It also scores 58.6% on SWE‑Bench Pro, 78.7% on OSWorld (computer operation), and 81.8% on CyberGym (cybersecurity).

Industry Voices

Cursor CEO Michael Truell says “GPT‑5.5 is noticeably smarter and more persistent than GPT‑5.4, staying on complex tasks longer without giving up.” An NVIDIA engineer adds, “Losing GPT‑5.5 feels like an amputation.”

Performance Without Compromise and Self‑Optimization

Despite being larger, GPT‑5.5 matches GPT‑5.4’s latency while delivering higher intelligence, thanks to a redesigned inference architecture running on NVIDIA GB200 and GB300 NVL72 systems. Remarkably, GPT‑5.5 helped engineers improve load‑balancing algorithms, boosting token generation speed by over 20%.

Research Capability and a New Mathematical Proof

On GeneBench (genomics) and BixBench (bioinformatics) GPT‑5.5 achieves leading scores. It even independently discovered a new proof for a Ramsey‑number problem, which was subsequently verified in the Lean proof assistant.

Real‑World Cases

Immunology professor Derya Unutmaz used GPT‑5.5 Pro to analyze a dataset of 62 samples and ~28,000 genes, producing a detailed report that would have taken his team months. Mathematics assistant professor Bartosz Naskręcki built an algebraic‑geometry visualization app from a single prompt in 11 minutes.

Impact on Daily Office Work

Over 85% of OpenAI employees now use Codex weekly. The legal team reviewed 24,771 K‑1 tax forms (71,637 pages) in two weeks less than the previous year. The finance team automated six months of market‑activity analysis, and the marketing team saved 5–10 hours per week on reporting.

Security Measures and Availability

Because GPT‑5.5’s abilities reach into cybersecurity and bio‑risk domains, OpenAI deployed its strongest safety measures to date, conducting specialized tests with internal and external red teams and gathering feedback from nearly 200 early partners. The company plans to use the capabilities for defense rather than restriction.

ChatGPT Plus, Pro, Business, and Enterprise users are gradually receiving the update; Codex already runs GPT‑5.5, and an API will be released pending additional safety review.