Six Weeks. One New Standard — OpenAI’s GPT-5.5 Arrives with Agentic Ambitions
OpenAI’s GPT-5.5 Is Out — and It Works More Like an Agent Than a Chatbot
Released April 23, 2026. Available now for Plus, Pro, Business, and Enterprise users.
OpenAI released GPT-5.5 on April 23, 2026, describing it as its smartest and most intuitive model to date. The release came roughly six weeks after GPT-5.4 — a pace that reflects just how fast the frontier AI field is moving. The model is available today for Plus, Pro, Business, and Enterprise users in ChatGPT and Codex.
What sets GPT-5.5 apart from its predecessor isn’t just raw performance. It’s the way the model operates. You can hand it a multi-part, open-ended task — the kind that would normally require step-by-step supervision — and it will plan, use tools, check its own work, and keep going until things are done. OpenAI President Greg Brockman called it “a big step towards more agentic and intuitive computing.” OpenAI is no longer positioning this as just a smarter chat model — it’s positioning it as a task runner.
The model was co-developed and tested with input from nearly 200 early-access partners before launch. Teams at OpenAI use it internally across finance, engineering, marketing, and operations — over 85% of OpenAI staff use Codex every week, the agentic coding platform now powered by GPT-5.5. GPT-5.5 Pro — a premium variant using parallel test-time computation — rolls out separately to Pro, Business, and Enterprise ChatGPT users.
GPT-5.5 at a Glance
Key figures from OpenAI’s April 2026 release
How GPT-5.5 Stacks Up
Select a benchmark to see GPT-5.5 results against competitors. Data sourced from OpenAI’s published system card and announcement.
Terminal-Bench 2.0 tests complex command-line workflows requiring planning, iteration, and tool coordination. Source: OpenAI GPT-5.5 announcement.
OSWorld-Verified measures whether a model can operate real desktop computer environments autonomously — clicking, navigating, and completing tasks without human guidance. Source: OpenAI GPT-5.5 announcement.
Tau2-bench Telecom tests complex multi-step customer-service workflows. Results run without prompt tuning, using GPT-4.1 as the user model. Source: OpenAI GPT-5.5 announcement.
SWE-Bench Pro evaluates real-world GitHub issue resolution in a single pass. Note: OpenAI stated that Anthropic reported signs of memorisation on a subset of SWE-Bench Pro problems in their own evaluations. Source: OpenAI GPT-5.5 announcement.
BrowseComp tests a model’s ability to track down hard-to-find information across the web. Score shown is for the GPT-5.5 Pro variant. Source: OpenAI GPT-5.5 announcement.
What People Who Used It Said
Statements sourced directly from OpenAI’s official GPT-5.5 release.
“GPT-5.5 is noticeably smarter and more persistent than GPT-5.4, with stronger coding performance and more reliable tool use. It stays on task for significantly longer without stopping early, which matters most for the complex, long-running work our users delegate to Cursor.”
Michael Truell — Co-founder & CEO, Cursor
“GPT-5.5 delivers the sustained performance required for execution-heavy work. It enables our teams to ship end-to-end features from natural language prompts, cut debug time from days to hours, and turn weeks of experimentation into overnight progress in complex codebases.”
Justin Boitano — VP of Enterprise AI, NVIDIA
“It’s incredibly energising to use OpenAI’s new GPT-5.5 model in our harness, have it reason over massive biochemical datasets to predict human drug outcomes, and then see it deliver significant accuracy gains on our hardest drug discovery evals.”
Brandon White — Co-Founder & CEO, Axiom Bio
“GPT-5.5 is the first coding model I’ve used that has serious conceptual clarity.”
Dan Shipper — Founder & CEO, Every
GPT-5.5 Availability — Tap Your Plan
Select a ChatGPT or Codex plan to see exactly what’s included.
GPT-5.5 API — What Developers Pay
Pricing confirmed by OpenAI as of April 24, 2026. Batch and Flex processing available at half the standard rate. Priority processing at 2.5× standard rate.
| Model | Input / 1M tokens | Cached Input | Output / 1M tokens | Context Window |
|---|---|---|---|---|
| gpt-5.5 | $5.00 | $0.50 | $30.00 | 1M tokens |
| gpt-5.5-pro | $30.00 | $3.00 | $180.00 | 1M tokens |
In Codex, GPT-5.5 runs with a 400K context window. Fast Mode delivers tokens 1.5× faster at 2.5× the cost. Source: OpenAI GPT-5.5 pricing page.
GPT-5.5 is co-designed for and served on NVIDIA GB200 NVL72 rack-scale systems — the same infrastructure central to next-generation AI inference. The race for AI coding dominance continues on multiple fronts: Microsoft passed on a $60 billion deal to acquire Cursor, while Xiaomi’s MiMo 2.5 Pro reached the top of open-source AI rankings on token efficiency — a metric GPT-5.5 also directly competes on. As AI becomes more embedded in enterprise workflows, questions about how companies handle employee data are also in the picture — Meta’s employee keystroke-tracking initiative for AI training data has already drawn scrutiny on that front. Meanwhile, leadership changes at major tech companies are adding another layer of change to the industry — including Tim Cook stepping down as Apple CEO, with John Ternus set to take over in September 2026.
What Was Covered
GPT-5.5 was released by OpenAI on April 23, 2026, for paid ChatGPT and Codex users. The model was positioned around agentic task execution — multi-step work involving planning, tool use, and self-checking, rather than single-turn responses.
Benchmarks covered in this piece included Terminal-Bench 2.0 (82.7%), OSWorld-Verified (78.7%), Tau2-bench Telecom (98.0%), SWE-Bench Pro (58.6%), GDPval (84.9%), and BrowseComp (90.1% for the Pro variant). The model matches GPT-5.4’s per-token latency in real-world serving while scoring higher across these evaluations, and uses fewer tokens for the same Codex tasks.
API pricing was set at $5 per million input tokens and $30 per million output tokens for gpt-5.5, and $30/$180 per million tokens for gpt-5.5-pro. OpenAI rated the model’s cybersecurity and biological capabilities as “High” under its Preparedness Framework — below “Critical” but above GPT-5.4’s rating. GPT-5.5 was co-designed for and served on NVIDIA GB200 NVL72 systems. The model is also available via Microsoft Foundry for enterprise teams building production AI agents.
