OpenAI Launches GPT-5.5, Touting Agentic Computing and 12M-Token Context Window

As generative artificial intelligence becomes more widespread, the global technology industry is moving beyond basic information retrieval and conversation toward an era of AI agents capable of performing work on people’s behalf. The shift will redefine how companies approach digital transformation and is expected to fundamentally change everyday office work for white-collar employees. It is also becoming a key battleground for major technology companies seeking commercial adoption.

U.S. technology giant OpenAI unveiled its new GPT-5.6 model family on July 16, 2026, including Sol, Terra and Luna. It also launched the ChatGPT Work agent, giving subscribers who pay $20 a month access to a new feature that can autonomously perform complex tasks across applications and files, significantly improving office automation efficiency.

OpenAI launched GPT-5.5 on April 23, 2026, and expanded Codex's role beyond software development to research, data analysis, document and spreadsheet creation, and software operation. The shift means AI agents can do more than assist with coding: they can complete longer, general-purpose computer workflows across multiple tools. The model became available through the API on April 24.

The UK AI Security Institute published an evaluation on April 30, 2026, finding that GPT-5.5 performed comparably to Anthropic's Claude Mythos Preview across 95 cybersecurity tasks spanning four difficulty levels. It also became the second model to complete an end-to-end simulated enterprise-network attack, a test estimated to take a human 20 hours. API pricing is $5 per million input tokens and $30 per million output tokens.

Since launching the GPT-5 series, OpenAI has continued to develop large language models beyond conversational tools into AI agents capable of executing multistep tasks. Released on March 5, 2026, GPT-5.4 is the company’s first general-purpose model with native computer-use capabilities. It can issue mouse and keyboard commands based on screenshots and complete tasks across websites, spreadsheets and documents, lowering the integration barrier for automating professional workflows.

OpenAI made GPT-5.4 Thinking available in ChatGPT, the API and Codex on the same day. API pricing is $2.50 per million input tokens and $15 per million output tokens. The model achieved an 83.0% win-or-tie rate on the GDPval professional-work benchmark. In Tool Search testing across 250 tasks, it reduced token usage by 47% without sacrificing accuracy.

OpenAI continues to compete in the GPT model race by expanding context windows and strengthening reasoning capabilities. If GPT-5.4 supports 1 million tokens, it could process large codebases, lengthy research materials and multiple documents in a single session, a capability that would be particularly important for researchers and enterprise AI applications.

As of July 20, 2026, reports said OpenAI was developing GPT-5.4, with plans to expand its context window to 1 million tokens and add an “extreme thinking mode” for complex reasoning tasks. The reports did not provide a release date, pricing or API rates, and OpenAI has yet to formally announce any details.