OpenAI Introduces GPT-55 Series AI Models With Improved Agentic Coding and Knowledge Work

Thursday, OpenAI announced the launch of a new major update to its GPT-5 series artificial intelligence (AI) models. taking it to GPT-5 with. The new models are said to have improved intent understanding, agentic coding and reasoning capabilities as well as the ability of 5 in its own right. AI giant San Francisco-based claims the models outperform Anthropic’s new Claude Opus 4. google – and Google’s Gemini 3 with 7). Across all disciplines, 1 Pro models are in the mix. The first is the introduction of these large language models (LLMs) to the company’s paid subscribers, and via the application programming interface (API)).

OpenAI Launches GPT-5.5

A post describing the GPT-5 was introduced and detailed by the company in a statement. AI 5 series models of . There are three variants the base model, GPT-5. GPT-5 & 5 thinking, and the paraphrasing. Using 5 Pro, paraphrasing is the most effective. GP 5 OpenAI is rolling out GPT-5 with . The GPT-5 is the only one in ChatGPT that has a thinking model to Plus, Pro, Business and Enterprise users. Users of Plus can’t use 5 Pro.

Codex The GPT-5 is a in CodeX. Go, Plus and Pro subscriber to 5 AI model is available for Business, Enterprise and Edu subscribers. The code platform will provide a context window of 400K tokens and ‘Face mode,’ with the ability to use an advanced programming platform. In a statement released by OpenAI, the AI models will soon be available via the Responses and Chat Completions APIs. Pricing is $5 (roughly Rs. A total of 471) per million input tokens and $30 (roughly Rs. Per million output tokens of , 2,828) were issued.

Until improvements, the models are heavily focused on agentic coding and intent understanding. OpenAI claims the GPT-5 is a . Five had 82 goals for 5. Terminal-Bench 2 7 percent . The 0 benchmark, which tests command-line workflows, is better than Anthropic and Google’s new models. Similarly, it scored 73 per cent of its s. SWE-Bench Pro, the company’s internal review of long-horizon coding projects is 1 percent on its website’.

Codex, powered by GPT-5, is alleged to be the real-world equivalent of . Can 5 manage the execution of code and refactoring, debugging, testing or validation for code implementation,. OpenAI says the model can also be able to reason by “ambiguous failures”, check assumptions with tools, and carry changes across the surrounding codebase.

Knowledge work is another key area that focuses on knowledge work. According to the model, it’s more intuitive than its predecessor and has better intent understanding as well as being a result of improved models. It can better measure the user’s real needs, find relevant information and review the output to produce something useful. In addition, these capabilities are also said to make the model better at scientific and technical research. GPT-5 . In OpenAI, 5 “can gather evidence and test assumptions,” read the results (and decide what to do next”).