OpenAI released GPT-5.5 on April 23, 2026, just six weeks after its predecessor GPT-5.4. The model — internally codenamed “Spud” — represents the company’s first fully retrained base model since GPT-4.5 and is built for complex, autonomous multi-step workflows. GPT-5.5 excels in agentic coding, knowledge work, and early scientific research, achieving state-of-the-art scores on several key benchmarks. However, it trails Anthropic’s Claude Opus 4.7 on real-world software engineering tasks and arrives with doubled API pricing, raising questions about total cost of ownership for enterprise adopters. The release is part of OpenAI’s broader push to unify ChatGPT, Codex, and its AI browser into a single “super app” for enterprise customers.
Key Overview
- Release date: April 23, 2026 — rolling out to Plus, Pro, Business, and Enterprise ChatGPT and Codex users
- Codename: Spud; first fully retrained base model since GPT-4.5
- Top benchmark scores: 82.7% on Terminal-Bench 2.0; 35.4% on FrontierMath; 84.9% on GDPval
- Key weakness: Trails Claude Opus 4.7 on SWE-Bench Pro (58.6% vs. 64.3%)
- Token efficiency: Approximately 40% fewer output tokens per Codex task compared to GPT-5.4
- API pricing: $5 per million input tokens and $30 per million output tokens — double GPT-5.4’s rates
- Super app vision: OpenAI aims to merge ChatGPT, Codex, and an AI browser into a unified enterprise platform
- Scale: ChatGPT now has over 900 million weekly active users and more than 50 million subscribers
Markets move fast; don’t get left behind. We’ve paired the Serrari Group Market Index with a curated Marketplace and a comprehensive Wealth Builder Platform to ensure you have the data—and the skills—to act on it.
A Rapid Release Cycle Reflects Intensifying Competition
OpenAI’s decision to ship GPT-5.5 just six weeks after debuting GPT-5.4 is emblematic of the breakneck pace currently defining the frontier AI industry. The release came only one week after Anthropic shipped Claude Opus 4.7, which reclaimed the lead in real-world software engineering benchmarks and introduced high-resolution vision capabilities. Google’s Gemini 3.1 Pro has also been competitive across multiple evaluations, keeping all three labs locked in a cycle of rapid iteration.
OpenAI president Greg Brockman described GPT-5.5 as a step toward more agentic and intuitive computing during a press briefing, emphasizing that the model can take unclear problems and determine what needs to happen next with far less guidance than its predecessors. Chief scientist Jakub Pachocki went further, telling reporters that the company sees significant improvements ahead and that, in his view, the last two years of model progress have been “surprisingly slow.”
This release cadence is not solely about technical superiority. As Fortune reported, OpenAI disclosed that ChatGPT now has more than 900 million weekly active users, over 50 million subscribers, 4 million active Codex users, and 9 million paying business users. These numbers are part of a deliberate effort to counter a growing narrative that the company has lost ground to Anthropic among enterprise customers.
Benchmark Performance: Dominance With Notable Gaps
GPT-5.5’s benchmark results paint a picture of broad strength with specific areas where competitors still lead. On Terminal-Bench 2.0, which evaluates an AI model’s ability to autonomously execute complex multi-step tasks in command-line environments — such as editing files, setting up servers, and installing tools — GPT-5.5 scored 82.7%. That is a substantial lead over Claude Opus 4.7 at 69.4% and Gemini 3.1 Pro at 68.5%.
On FrontierMath, a benchmark that measures performance on highly complex mathematical reasoning problems comparable to unsolved modern challenges, GPT-5.5 achieved 35.4%, outperforming Claude Opus 4.7 at 22.9% and Gemini 3.1 Pro at 16.7%. The model also performed well on business-oriented evaluations. On GDPval, which measures economically valuable knowledge work across 44 occupations, GPT-5.5 scored 84.9%. On OSWorld-Verified, which evaluates real-world desktop task execution, it reached 78.7% — a threshold that exceeds the estimated human average of 72.4%.
However, the picture is not uniformly dominant. On SWE-Bench Pro, which tests real-world GitHub issue resolution across multiple programming languages, Claude Opus 4.7 scored 64.3% compared to GPT-5.5’s 58.6%. This benchmark was notably absent from OpenAI’s official comparison materials. Independent analysis from Lushbinary characterized the competitive landscape as a split: GPT-5.5 leads planning-and-execution evaluations such as Terminal-Bench and Toolathlon, while Claude Opus 4.7 leads codebase-resolution evaluations like SWE-Bench Pro, SWE-Bench Verified, and CursorBench.
A broader head-to-head analysis across the ten benchmarks where both providers independently reported results found that Claude Opus 4.7 leads on six benchmarks, while GPT-5.5 leads on four. The two models are optimized for fundamentally different types of workflows rather than competing on a single axis.
Code Review: Higher Precision, Better Signal
Independent testing by CodeRabbit, an AI-driven code review platform, provided granular insight into GPT-5.5’s practical coding capabilities. In CodeRabbit’s curated review benchmark, GPT-5.5 raised the issue detection rate from 58.3% to 79.2% and improved precision from 27.9% to 40.6%, while generating a modestly higher number of comments — 75 versus the baseline’s 67.
The improvements extended to larger-scale real-world tests. Issue detection rates rose from 55.0% to 65.0%, precision improved from 11.6% to 13.2%, and total comments increased from 558 to 722. CodeRabbit noted that GPT-5.5 felt noticeably different in practice: quicker, leaner, and more direct than previous models, with a stronger bias toward small, workable changes rather than broad rewrites.
CodeRabbit’s evaluation also highlighted that GPT-5.5 was particularly strong in debugging-oriented reviews. When tasks involved access control, error handling, or API behavior, the model could isolate regressions, reject weak diagnoses, and point toward fixes that preserved intended behavior. This selectivity translates to reduced time spent sifting through irrelevant review comments and a greater likelihood that feedback highlights issues worth addressing.
Hands-on testing by independent developer Uygar Duzgun echoed these findings. He found that GPT-5.5 produced more disciplined patches, touching the right files, preserving existing coding styles, and stopping when the issue was actually fixed — rather than spiraling into unnecessary refactors. The model’s improvement, he argued, lies less in raw intelligence and more in restraint: a model that changes 800 lines to fix a 20-line issue may look impressive in a demo but becomes expensive in a real repository.
Context is everything. While you follow today’s updates, use the Serrari Group Market Index and Marketplace to spot emerging shifts. Need to sharpen your edge? Our Wealth Builder Platform turns these insights into a professional-grade strategy.
Token Efficiency: Doing More With Less
One of GPT-5.5’s most significant practical improvements is in token efficiency. OpenAI stated that the model uses significantly fewer tokens to complete the same Codex tasks as GPT-5.4, with internal measurements suggesting approximately 40% fewer output tokens per task. This is not a minor optimization. Token usage directly translates into cost for enterprise customers, and previous models often consumed tokens at each step of a planning-execution-review-retry cycle.
Independent testing by Artificial Analysis confirmed a large reduction in output token usage relative to GPT-5.4, finding that the effective API cost increase over GPT-5.4 was around 20% — far less than the 100% sticker price increase suggests. CodeRabbit similarly observed a clear reduction in token usage across its tests, with the model completing equivalent review tasks using fewer tokens than earlier versions.
OpenAI also noted that GPT-5.5 achieves its performance gains without sacrificing speed. The model matches GPT-5.4’s per-token latency in real-world serving, which is notable because larger, more capable models typically incur higher latency penalties. On the Artificial Analysis Coding Index, GPT-5.5 was assessed as delivering state-of-the-art intelligence at half the cost of competitive frontier coding models when efficiency is factored in.
Long-context performance also saw dramatic improvement. On MRCR v2, which tests how reliably a model can locate multiple pieces of information across very long texts, GPT-5.5 scored 74.0% at context lengths of 512K to 1M tokens — up from 36.6% for GPT-5.4.
API Pricing: A Doubled Sticker Price With Caveats
GPT-5.5’s API pricing is set at $5 per million input tokens and $30 per million output tokens, with a 1 million token context window. That is exactly double GPT-5.4’s rates of $2.50 and $15 respectively. GPT-5.5 Pro, designed for higher-accuracy work, is priced at $30 per million input tokens and $180 per million output tokens.
OpenAI offers tiered pricing to offset the increase. Batch and Flex processing are available at half the standard rate, while Priority processing runs at 2.5x. For prompts exceeding 272,000 input tokens, pricing scales to 2x input and 1.5x output for the full session.
The pricing increase has drawn scrutiny. As The Decoder reported, while the model tops many benchmarks, it still exhibits notable hallucination issues and the effective cost is roughly 20% higher than GPT-5.4 once token efficiency gains are factored in. Enterprise teams evaluating GPT-5.5 are advised to benchmark cost per completed task rather than comparing token prices in isolation.
For ChatGPT subscribers, GPT-5.5 Thinking is included in the Plus plan at $20 per month and the Pro plan at $200 per month. GPT-5.5 Pro is restricted to Pro, Business, and Enterprise tiers. Free-tier users do not currently have access.
Agentic Capabilities and the Super App Strategy
GPT-5.5 is designed around the concept of agentic computing — models that plan, act, check their own work, and continue across multiple steps rather than stopping at a first answer. OpenAI identifies four core areas of improvement: agentic coding, computer use, knowledge work, and early scientific research.
On Toolathlon, which measures the ability to complete complex workflows by integrating multiple software tools such as Notion, Slack, and databases, GPT-5.5 showed strong results. This benchmark reflects a shift in AI from chatbot-style interaction to autonomous agents capable of executing real tasks across platforms.
Mark Chen, OpenAI’s chief research officer, said during the press briefing that GPT-5.5 shows meaningful gains on scientific and technical research workflows and could help expert scientists make progress, noting particular promise in drug discovery. OpenAI also highlighted internal use cases: the company’s finance team used Codex with GPT-5.5 to review over 24,000 K-1 tax forms totaling more than 71,000 pages, accelerating the task by two weeks compared to the prior year.
The broader strategic vision is a unified service. OpenAI co-founders Sam Altman and Greg Brockman envision combining ChatGPT, Codex, and an AI browser into a single offering for enterprise customers — a concept the company internally frames as a “super app.” Enterprise adoption signals are already visible: over 10,000 NVIDIA employees across engineering, legal, marketing, and other functions reportedly have access to GPT-5.5-powered Codex, and OpenAI states that more than 85% of its own company uses Codex weekly across functions including finance, communications, and product management.
GitHub has also moved quickly. GPT-5.5 became generally available for GitHub Copilot on April 24, rolling out to Copilot Pro+, Business, and Enterprise users with a 7.5x premium request multiplier during its initial promotional pricing period.
Safety and Safeguards
OpenAI released GPT-5.5 with what it described as its strongest set of safeguards to date. The model underwent the company’s full suite of predeployment safety evaluations and Preparedness Framework, including targeted red-teaming for advanced cybersecurity and biology capabilities. OpenAI collected feedback from nearly 200 trusted early-access partners before the public release.
The company’s system card details evaluations of chain-of-thought controllability, a property that measures whether a model can be made to hide or reshape its reasoning in ways that would evade monitoring. GPT-5.5’s controllability was found to be lower than GPT-5.4 Thinking and GPT-5.2 Thinking — a result OpenAI considers positive, as it increases confidence in the reliability of reasoning monitoring.
The cybersecurity implications of increasingly capable models were a recurring theme in the press briefing. As CNBC reported, Mia Glaese, OpenAI’s VP of research, said the company has been iterating on its cyber safeguards for months with increasingly capable models and has refined what she called a durable approach to rolling out models safely. The safety conversation has been heightened industry-wide following Anthropic’s announcement of Claude Mythos, a model whose cybersecurity capabilities led Anthropic to restrict its general availability.
Where GPT-5.5 Fits in the Competitive Landscape
The April 2026 model releases represent one of the most competitive periods in AI history. Within a single week, Anthropic shipped Claude Opus 4.7 and OpenAI followed with GPT-5.5, while Google’s Gemini 3.1 Pro continued to perform competitively across multiple benchmarks.
The competitive dynamic is not a simple leaderboard. GPT-5.5 and Claude Opus 4.7 are optimized for fundamentally different workflows. GPT-5.5 excels at long-horizon agentic tasks that require planning, tool coordination, and autonomous execution. Claude Opus 4.7 dominates in precision-oriented software engineering tasks like resolving complex GitHub issues and multi-language codebase refactoring. For development teams, the practical guidance emerging from independent evaluations is that the right model depends on the specific workflow rather than aggregate benchmark rankings.
The six-week release cadence between GPT-5.4 and GPT-5.5 signals that OpenAI’s competitive strategy extends beyond model quality. The company appears focused on establishing platform lock-in before enterprise procurement cycles close, using its massive user base and expanding Codex ecosystem to embed itself as the default AI infrastructure for large organizations. Whether this velocity of iteration translates into sustained technical leadership or simply compresses the upgrade cycle for enterprise buyers remains to be seen.
Your financial future isn’t something you wait for—it’s something you build.
The real question is: when do you begin?
Move beyond simply staying informed.
Navigate the markets with clarity—track trends through the Serrari Group Market Index, uncover opportunities in the Serrari Marketplace, and build practical knowledge with our Curated Wealth Builder Platform.
Stay connected to what truly matters.
Get daily insights on macro trends and financial movements across Kenya, Africa, and global markets—delivered through the Serrari Newsletter.
Growth opens doors.
Advance your career through professional programs including ACCA, HESI A2, ATI TEAS 7 , HESI EXIT , NCLEX – RN and NCLEX – PN, Financial Literacy!🌟—designed to move you forward with confidence.
See where money is flowing—clearly and in real time.
Track Money Market Funds, Treasury Bills, Treasury Bonds, Green Bonds, and Fixed Deposits, alongside global and African indexes, key economic indicators, and the evolving Crypto and stablecoin landscape—all within Serrari’s Market Index.
