OpenAI Releases GPT-5.5 — A New Class of Intelligence Aimed Directly at Anthropic and Google

OpenAI just released GPT-5.5, rolling out today to Plus, Pro, Business, and Enterprise subscribers across ChatGPT and Codex. API access is coming very soon.

The model, internally codenamed Spud, follows GPT-5.4 by six weeks. OpenAI president Greg Brockman described it as "a new class of intelligence" and "a big step towards more agentic and intuitive computing."

GPT 5.5 is smarter than GPT 5.4, uses fewer tokens to complete the same tasks, and runs at the same speed. Getting more capability without sacrificing speed or driving up token cost is rare. Bigger models are almost always slower. GPT 5.5 is not.

What GPT-5.5 Is Actually Better At

OpenAI says the biggest gains are in four areas: agentic coding, computer use, knowledge work, and scientific research. Here is what each of those means in plain terms.

Agentic coding: Give it a messy, multi-step engineering problem and it plans the approach, uses tools, checks its own work, navigates ambiguity, and keeps going until the job is done. You do not manage every step. On Expert-SWE, an internal benchmark testing long-horizon coding tasks that take a human around 20 hours to complete, GPT 5.5 outperforms GPT 5.4. On SWE-Bench Pro, which grades real GitHub issue resolution, it scores 58.6%. Claude Opus 4.7 scores higher at 64.3%, though OpenAI disputes the methodology citing signs of memorization in Anthropic's benchmark runs.

Computer use: GPT 5.5 navigates software better than its predecessors because it understands intent more accurately. On Terminal-Bench 2.0, which tests complex command-line workflows that require planning and iterative tool use, GPT 5.5 scores 82.7%. Claude Opus 4.7 lands at 69.4% and Gemini 3.1 Pro at 68.5%. That is not a close race.

Knowledge work: On GDPval, a benchmark testing performance across 44 real occupations including finance, legal research, and product management, GPT 5.5 matches or beats industry professionals in 84.9% of comparisons. This is the benchmark that answers the question most people actually care about: can it do real office work?

Scientific research: OpenAI's chief research officer Mark Chen said GPT 5.5 shows meaningful gains on scientific and technical research workflows, including drug discovery. On OSWorld, which tests an AI's ability to operate a computer to complete real tasks across a full desktop environment, creating files, navigating apps, and building things, GPT 5.5 leads the field.

It Can Build a Game While You Watch

One of the most striking demos from the GPT 5.5 briefing: the model built a playable maze game in real time, applied textures, and modified gameplay mechanics mid-session, all from natural language instructions inside a live computer environment. This is not a code generator writing a script for you to run later. It is an AI operating a computer, building something, and iterating on it autonomously. That is what OSWorld measures and GPT-5.5 is currently the top scorer.

The Efficiency Story Nobody Is Talking About

GPT 5.5 completes the same Codex tasks with fewer tokens than GPT 5.4. Sam Altman addressed the pricing concern directly on X: even though the per-token rate is higher, the token efficiency gains mean most real-world runs cost the same or less than GPT 5.4.

On the Artificial Analysis Intelligence Index, which averages performance across a wide range of tasks, GPT 5.5 ranks as the most intelligent model available right now.

On BrowseComp, which tests a model's ability to track down hard-to-find information across the live web, GPT 5.5 Pro scores 90.1%, ahead of Gemini 3.1 Pro at 85.9%.

GPT-5.5 Pro

A separate GPT 5.5 Pro tier, for harder, higher-accuracy work, is rolling out to Pro, Business, and Enterprise users. This is the version scoring 90.1% on BrowseComp and delivering peak performance on complex reasoning benchmarks. API pricing for Pro: $30 per million input tokens and $180 per million output tokens.

Pricing

GPT 5.5 costs $5 per million input tokens and $30 per million output tokens. GPT 5.5 Pro costs $30 per million input tokens and $180 per million output tokens. For context, GPT-5.4 was priced at $2.50 input and $15 output per million tokens. The per-token rate is higher but OpenAI argues token efficiency on real workloads offsets the difference.

For comparison: Kimi K2.5 costs $0.44 input and $2.00 output per million tokens. Minimax M2.7 costs $0.30 and $1.20. GPT-5.5 sits in a different price tier entirely, built for enterprise use cases where accuracy and autonomy matter more than raw cost per token.

Who Can Use It Right Now

GPT-5.5 is live today in ChatGPT and Codex for Plus, Pro, Business, and Enterprise subscribers. GPT 5.5 Pro is rolling out to Pro, Business, and Enterprise. Free tier users do not get access. API is coming very soon.

You can try GPT-5.5 and compare it against every major model on Chatly right now from one workspace, no separate accounts needed.

The Bigger Picture

GPT-5.4 arrived two days after GPT-5.3. GPT-5.5 followed six weeks later. OpenAI's chief scientist Jakub Pachocki put it plainly: "I think the last two years have been surprisingly slow." The pace is accelerating, not slowing. OpenAI now has 900 million weekly active ChatGPT users, 50 million subscribers, and 9 million paying business users. GPT-5.5 is a direct response to competitive pressure from Anthropic's Claude Opus 4.7 and Mythos and a signal that OpenAI is building toward a single "superapp" combining ChatGPT, Codex, and an AI browser into one unified platform.