#OpenAIReleasesGPT-5.5

The Agentic AI Era Has Begun
On April 23, 2026, OpenAI announced the release of “GPT-5.5” to all paid users. Internally codenamed “Spud,” this version stands out with its “agentic” architecture that doesn’t just chat, but plans and completes assigned goals from start to finish. The launch, which came just one week after Anthropic’s Mythos model, showed that the AI race in 2026 has shifted into a higher gear.
What Is GPT-5.5 and What Can It Do?
OpenAI describes GPT-5.5 as “a new class of intelligence for real work.” Unlike previous versions, the model moves forward on vague, multi-step tasks without needing step-by-step instructions from the user. It plans, uses tools, checks its own work, runs code when needed, creates documents, conducts online research, and switches between different apps.

OpenAI co-founder Greg Brockman summed up the model’s positioning: “This isn’t an incremental update, it’s a new class of intelligence. It’s a big step toward more agentic and intuitive computing.”
Record-Breaking Results in Coding and Computer Use
GPT-5.5’s biggest leap is in agentic coding and computer use.
• Terminal-Bench 2.0: On this benchmark that tests complex command-line workflows, it reached 82.7% accuracy — a new state-of-the-art. • SWE-Bench Pro: In tests measuring real-world GitHub issue resolution, it scored 58.6%, solving more tasks end-to-end in a single pass than previous models. • OSWorld-Verified: On the benchmark that measures whether a model can operate a computer independently, it achieved 78.7%. GPT-5.4 scored 75%, while Anthropic Opus 4.7 scored 78%.
It does all this while using fewer tokens than GPT-5.4. According to Artificial Analysis’s Coding Index, GPT-5.5 delivers frontier-level intelligence at roughly half the cost of competing models.
Who Has Access and What’s the Pricing?
GPT-5.5 went live for ChatGPT Plus, Pro, Business, and Enterprise users on launch day. API access opened on April 24. A larger “GPT-5.5 Pro” version is also available with the $100/month Pro plan and to Business/Enterprise accounts.

There is no access yet for free-tier users. OpenAI said the API was opened with a one-day delay due to the need for “different safety measures.”
Why “Spud”? The Development Process
The model’s codename is known as “Spud.” Pre-training was completed in March 2026. Sam Altman described the model to staff as “a very powerful model that could actually speed up the economy.” Greg Brockman said Spud is “the product of two years of research” and represents “not an incremental step, but a significant shift in how we think about model development.”
Enterprise Use: Claims of 10 Hours Saved
Teams with early access reported they could review thousands of additional documents, catch errors in quickly written code, and save up to 10 hours of work per week. The model excels at office tasks like creating documents, spreadsheets, presentations, data analysis, and running software.
Competition: Scores vs. Anthropic and Google
GPT-5.5 outperformed Anthropic’s Opus 4.7 and Google’s Gemini 3.1 Pro in coding benchmarks. It pulled ahead especially in agentic tasks — scenarios requiring multi-step planning and tool use. Compared to GPT-5.4, it’s positioned as a “faster, sharper thinker for fewer tokens.”
Safety and “Agentic” Risks
OpenAI says GPT-5.5 represents the first step toward “AI systems that can complete complex, multistep tasks on a computer without human guidance.” For that reason, the model, defined as “agentic AI,” ships with stricter safety guardrails. The company emphasizes that the model can correct its own mistakes and navigate ambiguity.
What’s Next: GPT-6 Is Coming
Industry sources see GPT-5.5 as a bridge to GPT-6, which is expected to bring “infinite context,” advanced memory management, and a new agent mode. There’s also talk that OpenAI is working on a “super app” integrating ChatGPT, Codex, and the Atlas browser.

Summary: #OpenAIReleasesGPT-5.5 marks the transition of AI from the “Q&A” phase to the “get the job done end-to-end” phase. With an 82.7% Terminal-Bench score, frontier performance at half the cost, and a promise of 10 hours saved per week, GPT-5.5 sets a new standard in coding, research, and enterprise automation. The agentic era has officially begun, and the competition is now built not just on “better answers,” but on “completing more work.”