OpenAI's Full Suite Launch: GPT-5.6 Trio Leads the Way, Work and Sites Follow Suit

On July 9, OpenAI laid all its cards on the table for the entire year.

The full GPT-5.6 series was released, ChatGPT Work's cross-app Agent officially launched, and ChatGPT Sites turned conversations into websites—three products dropping on the same day was no coincidence.

💡 What You Will Learn

OpenAI's Full Suite Launch: GPT-5.6 Trio Leads the Way, Work and Sites Follow Suit On July 9, OpenAI laid all its cards on the table for the entire year. The full GPT-5.6 series was released, ChatGP

📜 Table of Contents

• Three Pieces, Three Layers
• GPT-5.6: Not One Model, But Three
• ChatGPT Work: The Qualitative Leap from Chatting to Working
• ChatGPT Sites: Your Conversation Is a Website
• The Chemical Reaction of the Trio

OpenAI's Full Suite Launch: GPT-5.6 Trio Leads, Work and Sites Follow

On July 9th, OpenAI laid all its cards on the table for the year.

The GPT-5.6 series launched across the board, ChatGPT Work's cross-app Agent went live, and ChatGPT Sites turned conversations into websites—three products dropping on the same day isn't a coincidence; it's a complete strategic loop.

This isn't "one company releasing three new products." It's a platform simultaneously rolling out three layers of capability: the model layer, the Agent layer, and the application layer. Stitched together, these three pieces reveal OpenAI's true ambition—shifting from "you come to AI to chat" to "AI does the work for you."

Three Pieces, Three Layers

Before diving into each one, let's look at the big picture:

This isn't a product line—it's a tech stack. The foundation provides compute, the Agent layer provides execution capability, and the delivery layer provides output. The combined effect of these three layers far exceeds the sum of individual products.

GPT-5.6: Not One Model, But Three

This time, OpenAI didn't follow the old "one model to rule them all" path. The GPT-5.6 trio is directly tiered by price point and use case:

The benchmark data tells a compelling story. Sol scored 53.6 on Agents' Last Exam—a test simulating long-horizon workflows across 55 professional domains—and with just medium reasoning enabled, it beats Fable 5 by 11.4 points at a quarter of the cost. Terra and Luna go even further, outperforming Fable 5 at one-sixteenth the cost.

On coding capability, Sol scored 80 on the Coding Agent Index, 2.8 points above Fable 5, while using half the output tokens, half the time, and a third of the cost. Altman himself noted a 54% improvement in token efficiency.

The key isn't the specs themselves—it's the strategy: OpenAI is no longer competing on "whose model is strongest," but on "who gives you the best option at every price point." Sol isn't benchmarked against a specific rival; it's about "you want to do something big? Here are 4 parallel Agents." Terra and Luna push cost-efficiency to the limit, leaving competitors with no room to fight on price.

ChatGPT Work: The Qualitative Leap from Chatting to Working

GPT-5.6 is the engine, but ChatGPT Work is the real "main course" for users in this full suite.

This isn't a simple "upgraded chat." It's an Agent that works across applications—opening your Google Drive, email, Slack, Teams, SharePoint, and Salesforce, reading files, querying data, building spreadsheets, writing slides, and generating documents on its own. You assign a task; it breaks it down into steps, executes them one by one, and can even pause mid-way to check in with you for feedback.

Zapier's head of marketing used it to build a system that automatically reviews thousands of leads daily, running data flows across CRM, email, and sales tools, uncovering seven-figure potential revenue. Virgin Atlantic's product lead used it to compress weeks of analysis into hours. An NVIDIA manager automated GTC conference prep with ChatGPT Work, eliminating the manual Excel workflows that previously consumed 40% of pre-event time.

Internally at OpenAI, the finance team used ChatGPT Work to shrink month-end close and forecasting from days to hours—finding data, filling tables, reconciling, building slides, and verifying results, all automated. The sales team completed a POC proposal that normally took weeks in under 24 hours.

Behind this is a key number: over 5 million people use Codex weekly, and 1 million of them aren't developers. ChatGPT Work extends Codex from "writing code" to "doing all work."

ChatGPT Sites: Your Conversation Is a Website

The third product looks the lightest, but its strategic weight is significant.

ChatGPT Sites lets you describe an idea, and it generates a publishable website for you. This isn't a toy "personal homepage generator"—built on GPT-5.6's Agent capabilities, Sites can understand your needs, handle design, write code, and deploy. From idea to launch, no IDE required.

This is arguably the most "ruthless" piece of OpenAI's trio: it lowers the delivery barrier to "as long as you can talk."

The Chemical Reaction of the Trio

Individually, each product has its own highlights. But together, the chemical reaction is what deserves attention:

GPT-5.6 provides the underlying reasoning power for Work and Sites. Work's ability to break down and execute steps across apps relies on GPT-5.6's Agent capabilities. Sites' ability to generate websites from conversation relies on the same model. The trio shares one brain.

Work provides the "execution" middle layer for Sites. Sites isn't just generating static pages—behind it, Work's Agent is decomposing requirements, designing architecture, and writing code. Without Work, Sites is just a template generator.

Sites provides the delivery outlet for the entire system. No matter how strong the model or how capable the Agent, everything ultimately needs to land in something users can get their hands on. Sites is that "last mile"—from idea to publishable result, no IDE required.

In the broader industry context, this combination punch becomes even clearer. Over the past two years, AI companies have been competing on "whose model is stronger." But starting with GPT-5.6, OpenAI's competitive dimension has shifted—it's no longer fighting on model specs alone, but on "you have a model; I have model + Agent + application" full-stack capability.

This is a signal: AI competition has moved from the "model arms race" to the "platform ecosystem war." Whoever builds the complete chain from foundation to application layer first holds the initiative in the next phase.