OpenAI's GPT-5.5 is built to work without handholding, and that is the point
OpenAI on Thursday released GPT-5.5, its first fully retrained base model since GPT-4.5, in a direct bid to reclaim enterprise ground from Anthropic. Codenamed "Spud," the model is designed to take messy, multi-part instructions and execute them autonomously across coding, computer use, knowledge work, and early scientific research. The company has been in what internal sources described as a "Code Red" state since December 2025, watching Anthropic's annual recurring revenue climb from $9 billion to $30 billion while its own B2B positioning eroded. GPT-5.5 is OpenAI's answer: a model that does not just answer questions but finishes jobs.
"We've expanded browser use so Codex can interact with web apps, and test flows, click through pages, capture screenshots, and iterate on what it sees until it completes the…" OpenAI Developers posted on X. The benchmark numbers back up the ambition. On Terminal-Bench 2.0, which tests complex command-line workflows requiring planning and tool coordination across multiple steps, GPT-5.5 scored 82.7%, up from 75.1% for GPT-5.4 released seven weeks ago. On SWE-Bench Pro, which evaluates real-world GitHub issue resolution across four programming languages, it hit 58.6%, solving more than half of issues in a single pass. On GDPval, a benchmark measuring professional task performance across 44 occupations (research, spreadsheet creation, document drafting), GPT-5.5 scored 84.9%. On OSWorld-Verified, which tests autonomous computer operation without human input, it reached 78.7%. On Tau2-bench Telecom, it scored 98% without any prompt tuning.
OpenAI says the efficiency gains are commercially significant for enterprise customers who pay per token but care about cost per completed task. The model uses "significantly fewer" tokens than GPT-5.4 to finish equivalent work in Codex while matching its predecessor's per-token latency meaning no speed tradeoff for higher intelligence.
API pricing reflects the upgrade at $5 per million input tokens and $30 per million output tokens, roughly double GPT-5.4 rates. But OpenAI argues the net cost per task drops because the model needs fewer tokens and fewer retries.
Absent from launch day: API access. OpenAI says API deployments require different safeguards and that it is working with partners on security requirements for serving at scale. The company promises access "very soon" but has not set a date.
GPT-5.5 rolls out today to Plus, Pro, Business, and Enterprise users in ChatGPT and Codex with a 400K context window in Codex (1 million when API launches). GPT-5.5 Pro with extended reasoning goes to Pro, Business, and Enterprise subscribers only. The safety posture is more cautious than previous releases: OpenAI evaluated the model across its full suite of preparedness frameworks with internal and external red-teamers, added targeted testing for cybersecurity and biology capabilities (including stricter classifiers for potential cyber risk), and collected feedback from nearly 200 trusted early-access partners before launch.
GPT-5.2 Thinking will remain available as a legacy option until June 5 before being retired. The release lands just days before Elon Musk's trial against OpenAI executives Sam Altman and Greg Brockman begins Monday in Oakland federal court.
