TL;DR GPT-5.5, codenamed "Spud," launched on April 23, 2026, and is available now in ChatGPT and Codex for all paid subscribers. It leads in agentic coding with an 82.7% score on Terminal-Bench 2.0 and an 85.0% on ARC-AGI-2 for abstract reasoning, both well ahead of Claude Opus 4.7 and Gemini 3.1 Pro. API access is not yet live, and the per-token price has doubled compared to GPT-5.4.

OpenAI released GPT-5.5 on April 23, 2026. The model, codenamed "Spud" internally, is now live in ChatGPT and Codex for paid subscribers.

This is OpenAI's first fully retrained base model since GPT-4.5. Every version between GPT-5 and GPT-5.4 was built on the same underlying architecture, refined and optimized, but never rebuilt from scratch. GPT-5.5 breaks that pattern. It is a new foundation, and OpenAI is treating it as one, positioning the model as the technical backbone for its planned "super app" that would unify ChatGPT, Codex, and a dedicated browser into a single desktop product.

Two variants shipped: GPT-5.5, the standard model available to all paid tiers, and GPT-5.5 Pro, a higher accuracy variant restricted to Pro, Business, and Enterprise users.

API access has not launched yet. OpenAI says it is coming "very soon" but requires additional cybersecurity safeguards before deployment at scale.

What is new in GPT-5.5: key features and performance improvements

GPT-5.5 is an agentic model. That means it is designed to handle complex, multi-step tasks on its own, moving across tools, checking its own work, and continuing until a job is done. Rather than requiring users to break problems down into granular instructions, GPT-5.5 interprets higher-level goals and figures out what needs to happen next.

The strongest improvements land in four areas: agentic coding, computer use, knowledge work, and early scientific research. OpenAI also reports that a customized version of GPT-5.5 contributed to the discovery of a new mathematical proof related to Ramsey numbers in combinatorics.

On the infrastructure side, GPT-5.5 matches GPT-5.4 in per-token latency despite being a larger model. It uses roughly 40% fewer output tokens to complete the same Codex tasks.

OpenAI attributes the speed gains to custom heuristic algorithms that partition GPU workloads based on real production traffic patterns, increasing token generation speeds by over 20%.

The model supports a 1 million token context window. Crucially, this is the first GPT series model where performance holds steady across the full length of that window. Previous models technically supported long contexts but degraded in quality past 128K tokens.

On OpenAI's needle-in-a-haystack benchmark (OpenAI MRCR v2), GPT-5.5 scored 87.5% at the 128K to 256K range and 74.0% at the 512K to 1M range, compared to GPT-5.4's 79.3% and 36.6%, respectively.

GPT-5.5 vs the competition: benchmark comparison

The benchmarks below are drawn from OpenAI's published evaluation tables. They tell a specific story: GPT-5.5 leads in agentic coding, computer use, mathematics, and cybersecurity, while Claude Opus 4.7 holds the edge in software engineering and tool use, and Gemini 3.1 Pro remains competitive in academic reasoning.

Benchmark comparison (GPT-5.5 vs competitors) at a glance

The benchmarks below are drawn from OpenAI's published evaluation tables. They show where GPT-5.5 leads and where competitors remain strong.

Coding

Eval	GPT-5.5	GPT-5.4	GPT-5.5 Pro	GPT-5.4 Pro	Claude Opus 4.7	Gemini 3.1 Pro
SWE-Bench Pro (Public)*	58.6%	57.7%	-	-	64.3%	54.2%
Terminal-Bench 2.0	82.7%	75.1%	-	-	69.4%	68.5%
Expert-SWE (Internal)	73.1%	68.5%	-	-	-	-

*Labs have noted evidence of memorization on this eval

Professional

Eval	GPT-5.5	GPT-5.4	GPT-5.5 Pro	GPT-5.4 Pro	Claude Opus 4.7	Gemini 3.1 Pro
GDPval (wins or ties)	84.9%	83.0%	82.3%	82.0%	80.3%	67.3%
FinanceAgent v1.1	60.0%	56.0%	-	61.5%	64.4%	59.7%
Investment Banking Modeling Tasks	88.5%	87.3%	88.6%	83.6%	-	-
OfficeQA Pro	54.1%	53.2%	-	-	43.6%	18.1%

Computer Use and Vision

Eval	GPT-5.5	GPT-5.4	GPT-5.5 Pro	GPT-5.4 Pro	Claude Opus 4.7	Gemini 3.1 Pro
OSWorld-Verified	78.7%	75.0%	-	-	78.0%	-
MMMU Pro (no tools)	81.2%	81.2%	-	-	-	80.5%
MMMU Pro (with tools)	83.2%	82.1%	-	-	-	-

GPT-5.5 vs Claude Opus 4.7 vs Gemini 3.1 Pro: coding benchmarks

Eval	GPT-5.5	GPT-5.4	GPT-5.5 Pro	Claude Opus 4.7	Gemini 3.1 Pro
Terminal-Bench 2.0	82.7%	75.1%	—	69.4%	68.5%
SWE-Bench Pro (Public)*	58.6%	57.7%	—	64.3%	54.2%
Expert-SWE (Internal)	73.1%	68.5%	—	—	—

* Labs have noted evidence of memorization on this eval

The 82.7% on Terminal-Bench 2.0 is the headline number. This benchmark tests complex command-line workflows requiring planning, iteration, and tool coordination.

GPT-5.5 leads every competitor by more than 13 percentage points. The 73.1% on

Expert-SWE is equally significant: this is OpenAI's internal benchmark for coding tasks with a median estimated human completion time of 20 hours, and it represents a 4.6 point jump over GPT-5.4.

Claude Opus 4.7 still leads on SWE-Bench Pro at 64.3%, which evaluates real GitHub issue resolution. That gap (5.7 points) is worth noting for teams whose primary use case is automated pull request work.

GPT-5.5 vs Claude Opus 4.7 vs Gemini 3.1 Pro: academic and math benchmarks

Eval	GPT-5.5	GPT-5.4	GPT-5.5 Pro	Claude Opus 4.7	Gemini 3.1 Pro
GeneBench	25.0%	19.0%	33.2%	—	—
FrontierMath Tier 1–3	51.7%	47.6%	52.4%	43.8%	36.9%
FrontierMath Tier 4	35.4%	27.1%	39.6%	22.9%	16.7%
BixBench	80.5%	74.0%	—	—	—
GPQA Diamond	93.6%	92.8%	—	94.2%	94.3%
Humanity's Last Exam (no tools)	41.4%	39.8%	43.1%	46.9%	44.4%
Humanity's Last Exam (with tools)	52.2%	52.1%	57.2%	54.7%	51.4%
ARC-AGI-1 (Verified)	95.0%	93.7%	—	93.5%	98.0%
ARC-AGI-2 (Verified)	85.0%	73.3%	—	75.8%	77.1%

GPT-5.5's 85.0% on ARC-AGI-2, a benchmark for abstract reasoning, is a 9.2 point lead over Claude Opus 4.7 and an 11.7 point jump from GPT-5.4.

On FrontierMath Tier 4, a set of postdoctoral-level math problems that can take human experts days to solve, GPT-5.5 Pro scored 39.6%, nearly double Opus 4.7's 22.9%.

On pure academic reasoning without tools (Humanity's Last Exam), Claude Opus 4.7 retains the lead at 46.9%.

GPT-5.5 vs Claude Opus 4.7 vs Gemini 3.1 Pro: professional and knowledge work benchmarks

Eval	GPT-5.5	GPT-5.4	GPT-5.5 Pro	Claude Opus 4.7	Gemini 3.1 Pro
GDPval (wins or ties)	84.9%	83.0%	82.3%	80.3%	67.3%
FinanceAgent v1.1	60.0%	56.0%	—	64.4%	59.7%
Investment Banking (Internal)	88.5%	87.3%	88.6%	—	—
OfficeQA Pro	54.1%	53.2%	—	43.6%	18.1%

GDPval measures practical knowledge work across 44 occupations. GPT-5.5's 84.9% is a modest improvement over GPT-5.4's 83.0%, which suggests that for everyday professional tasks the leap is less dramatic than the coding numbers imply.

The OfficeQA Pro score (54.1% vs 43.6% for Claude Opus 4.7 and 18.1% for Gemini 3.1 Pro) is a clearer separation, relevant for teams evaluating models for document and office productivity tasks.

GPT-5.5 vs Claude Opus 4.7 vs Gemini 3.1 Pro: computer use and vision benchmarks

Eval	GPT-5.5	GPT-5.4	Claude Opus 4.7	Gemini 3.1 Pro
OSWorld-Verified	78.7%	75.0%	78.0%	—
MMMU Pro (no tools)	81.2%	81.2%	—	80.5%
MMMU Pro (with tools)	83.2%	82.1%	—	—

OSWorld-Verified tests full desktop computer use. GPT-5.5's 78.7% narrowly edges Claude Opus 4.7's 78.0%, the first time OpenAI's mainline model has pulled ahead of Anthropic on this benchmark.

GPT-5.5 vs Claude Opus 4.7 vs Gemini 3.1 Pro: tool use benchmarks

Eval	GPT-5.5	GPT-5.4	GPT-5.5 Pro	Claude Opus 4.7	Gemini 3.1 Pro
BrowseComp	84.4%	82.7%	90.1%	79.3%	85.9%
MCP Atlas**	75.3%	70.6%	—	79.1%	78.2%
Toolathlon	55.6%	54.6%	—	—	48.8%
Tau2-Bench Telecom*	98.0%	92.8%	—	—	—

* Tau2-Bench Telecom: results use original prompts (no prompt adjustment)
** MCP Atlas: results from Scale AI after April 2026 update

Tau2-Bench Telecom: results for GPT-5.5 and GPT-5.4 with original prompts, no prompt adjustment.

On MCP Atlas, a tool-use benchmark, both Claude Opus 4.7 and Gemini 3.1 Pro outperform GPT-5.5. On BrowseComp, a web research benchmark, Gemini 3.1 Pro edges ahead of GPT-5.5's base model, though GPT-5.5 Pro leads at 90.1%.

The Tau2-Bench Telecom score of 98.0% is a standout, up from 92.8% for GPT-5.4.

GPT-5.5 vs Claude Opus 4.7: cybersecurity benchmarks

Eval	GPT-5.5	GPT-5.4	Claude Opus 4.7
Capture-the-Flags (Internal)****	88.1%	83.7%	—
CyberGym	81.8%	79.0%	73.1%

**** An expansion of the hardest CTFs used in system cards with additional hard challenges.

OpenAI classified GPT-5.5's cybersecurity capabilities as "High" under its Preparedness Framework. The CyberGym score of 81.8% leads Claude Opus 4.7 by 8.7 points.

This is also why OpenAI introduced Trusted Access for Cyber, a program that gives verified security professionals expanded access with fewer refusals.

One independent concern across all categories: Artificial Analysis reported an 86% hallucination rate for GPT-5.5 on their AA-Omniscience evaluation, compared to 36% for Claude Opus 4.7 and 50% for Gemini 3.1 Pro.

The model scored highest on raw knowledge accuracy, but its tendency to answer confidently when uncertain was flagged as a significant weakness.

GPT-5.5 vs GPT-5.4: Is it worth the upgrade?

The per-token API price has doubled. GPT-5.5 will cost $5 per million input tokens and $30 per million output tokens, compared to GPT-5.4's $2.50 and $15. GPT-5.5 Pro will cost $30 per million input tokens and $180 per million output tokens.

OpenAI's argument is that the efficiency gains offset the increase because the model needs fewer tokens to complete the same work. Whether that math holds depends on your specific workloads.

How to access GPT-5.5

To access GPT-5.5, users must have an active paid ChatGPT subscription. The model is available now to Plus ($20 per month), Pro ($200 per month), Business, and Enterprise users in ChatGPT and Codex.

GPT-5.5 Pro is available to Pro, Business, and Enterprise users only. Free-tier users do not have access. API access has not launched yet but OpenAI has confirmed it is coming soon, pending completion of additional cybersecurity safeguards.

Codex users across Plus, Pro, Business, Enterprise, Edu, and Go plans can access GPT-5.5 with a 400K context window.

For teams building with the API, the wait may be brief. OpenAI stated that "API deployments require different safeguards" and that they are working closely with partners on security requirements for serving at scale.

What GPT-5.5 means for teams building AI products

GPT-5.5 showcases what models can do autonomously. The improvements in agentic coding, computer use, and long-context handling point toward a future where AI manages entire workflows rather than responding to individual prompts.

But as model outputs grow richer and more structured, the presentation layer needs to keep pace. A model that can analyze a dataset, write a report, and generate a spreadsheet is only as useful as the interface that delivers those results to end users. This is the problem Generative UI solves.

Explore Generative UI with Thesys

At Thesys, our C1 API turns LLM responses into live, adaptive interfaces in real time: charts, forms, tables, reports, and layouts that render as the model works. When a model like GPT-5.5 produces multi-step, multi-format outputs, C1 ensures the intelligence translates into something users can see, interact with, and act on, without teams rebuilding their frontend for every new model release.

To summarize, GPT-5.5 is a capable model with real improvements in the areas that matter most for enterprise AI. It also comes with higher costs, an uneven benchmark profile, and a hallucination concern that independent evaluators have flagged. For developers and businesses evaluating it, the question is not whether GPT-5.5 is the best model in every category. It is whether the specific gains it offers align with the work they need it to do.

FAQs about GPT-5.5

Is GPT-5.5 free?

No. GPT-5.5 is available only to paid ChatGPT subscribers. Plus users pay $20 per month. Pro users pay $200 per month and also get GPT-5.5 Pro. The $180 gap between those tiers is significant. The Pro plan is designed for independent developers, researchers, and high-output power users who need extended reasoning, higher rate limits, and access to the most accurate model variant for correctness-critical work in fields like law, data science, and financial modeling. For most users, the Plus tier at $20 provides full access to GPT-5.5's core capabilities. There is no free-tier access to this model.

Is GPT-5.5 a flop?

The benchmark picture is genuinely mixed. GPT-5.5 leads across 14 benchmarks in publicly available models, compared to 4 for Claude Opus 4.7 and 2 for Gemini 3.1 Pro. But the hallucination concern is real, the GDPval improvement over GPT-5.4 is marginal, and Claude Opus 4.7 still leads in software engineering. For agentic coding and computer use, GPT-5.5 is currently the strongest available option. For pure reasoning without tools, other models remain competitive.

What is the smartest GPT model right now?

GPT-5.5 Pro is OpenAI's most capable model as of April 2026, designed for correctness-critical tasks in legal research, data science, and advanced analytics. It is available only on Pro, Business, and Enterprise plans.

Can GPT-5.5 be used for coding?

Yes. Coding is GPT-5.5's strongest area. It scored 82.7% on Terminal-Bench 2.0 and 73.1% on Expert-SWE, OpenAI's benchmark for tasks with a 20-hour human completion time. It is available in Codex for writing, debugging, and reviewing code, and can handle multi-file refactoring and iterative debugging autonomously.

Related articles

May 2026 Security Incident

Why We're Open Sourcing OpenUI

AI Copilots: A Complete Guide to Intelligent Assistance

What Are AI Agents: Types, Capabilities, and Real-World Use

ChatKit: What It Is, What It Isn’t, and Where C1 Fits In

PowerPoint Automation Reimagined with C1 Artifacts API

The Future of AI Interfaces is Generative

MCP UI: Making AI Apps Interactive

What Is AG-UI and Why It Matters for AI-Native User Interfaces

How to design AI-Native Conversational Interfaces : From Templates to Generative UI

GPT 5 vs. GPT 4.1

How to Build Generative UI Applications

Implementing Generative Analytics with Thesys and MCP

Evolution of Analytics: From Static Dashboards to Generative UI

Why Generating Code for Generative UI is a bad idea

Building the First Generative UI API: Technical Architecture and Design Decisions Behind C1

How we evaluate LLMs for Generative UI

Generative UI vs Prompt to UI vs Prompt to Design

Generative UI: What It Is and Why It Matters