5 min readNew DelhiUpdated: Apr 24, 2026 09:03 AM IST
OpenAI, on Thursday, April 23, introduced its latest and most advanced AI model – GPT-5.5. The Sam Altman-led company claims that the model is its smartest and most intuitive yet. It is being dubbed as the next phase in how work gets done. The company claims that the model is being developed to carry out complex real-world work.
GPT-5.5 is the latest large language model (LLM) in OpenAI’s GPT series, and it promises greater speed, accuracy, and efficiency. The company said that the model builds on its predecessor, GPT-5; however, it focuses more heavily on real-world applications such as research, data analysis, coding, document-heavy workflows, etc. While earlier versions followed step-by-step prompting from a user to carry out tasks, GPT-5.5 is claimed to be built to understand tasks earlier on, requires less guidance, and is able to take on tasks through completion. Along with the base model, OpenAI is also introducing GPT-5.5 Pro aimed at more demanding use cases.
Notably, a key shift here is the emphasis on practical and professional use. GPT-5.5 is optimised for tasks such as writing and debugging code; analysing large datasets; conducting research and synthesising information; creating documents, spreadsheets, and reports; and using tools and software to complete tasks. According to OpenAI, the model is particularly strong when it comes to execution-heavy tasks that need multiple steps, iteration, and decision-making along the way.
The new model also introduces some improvements in how it thinks through problems. The feature called ‘GPT-5.5 Thinking’ lets the model offer faster and more concise answers while still managing difficult tasks. Based on the company’s official release, when compared to earlier models, GPT-5.5 can produce more structured and relevant outputs, uses fewer tokens, better understands user intent, and requires fewer follow-up corrections. Essentially, these improvements make the model more suited for practical everyday professional use.
Another major highlight of GPT-5.5 is its ability to perform like an agent, meaning it can operate as a system that can plan, act, and adapt over time. Similar to agentic systems, the model can break down complex problems into steps, use tools and external systems efficiently, check its own work and make corrections, and continue working until a task is complete. This simply means the users can rely less on micromanaging prompts and instead give high-level instructions.
When it comes to performance, the new model reportedly achieves state-of-the-art performance across several benchmarks that are designed to test real-world knowledge work. On GDPval, which measures performance across 44 professions, it scored 84.9 per cent, and on OSWorld-verified, which tests its ability in various computer environments, it secured 78.7 per cent. Meanwhile, on Tau2-bench telecom, a benchmark focused on customer service workflows, the model secured 98.0 per cent. Besides, it also obtained strong scores in finance, office tasks, and investment modelling.
GPT-5.5 has also shown significant gains in scientific research. The model can assist full research processes by exploring ideas, gathering and analysing data, testing assumptions, interpreting results, and even suggesting next steps. In some cases, it was reportedly used to analyse complex biological datasets and generate detailed research reports, something that would take months to prepare. Meanwhile, on the safety front, OpenAI claims to have strengthened its guardrails for GPT5.5. The company has introduced initiatives such as a biosecurity bug bounty programme to test and improve safeguards against misuse.
Story continues below this ad
GPT-5.5 is OpenAI’s biggest announcement this week. On Wednesday, the AI powerhouse had introduced Workspace Agents, signalling its pivot to the agentic AI space for enterprise. The latest offering from the company reflects the ongoing broader shift in AI development which has now moved from systems that generate answers to systems that complete tasks. In a way, GPT-5.5 is not about incremental improvements but more about changing how people use AI in everyday work.
GPT-5.5 is rolling out across ChatGPT and Codex for Plus, Pro, Business, and Enterprise users, with Pro versions for higher accuracy tasks. Thinking mode is included in ChatGPT. Codex offers a 400K context window and a faster mode that generates responses 1.5x quicker at 2.5x higher cost. API access is coming soon.
