OpenAI just had one of its busiest product weeks of the year. In the span of a few days, it shipped a new flagship model family, a full agent product, a brand-new desktop app, a public website builder, and a next-generation voice mode — all landing directly inside ChatGPT. Here's a clear breakdown of what actually launched, what it replaces, and why each piece matters.
Instead of shipping one flagship model, OpenAI introduced GPT-5.6 as a family of three, each tuned for a different job:
OpenAI's own benchmarks claim Sol sets a new state of the art on coding-agent evaluations while using under half the tokens and taking under half the time of comparable models.
GPT-5.6 is now live across ChatGPT, Codex, and the API, with Sol priced at $5 input / $30 output per million tokens, Terra at $2.50 / $15, and Luna at $1 / $6.
🔗 OpenAI release notes: GPT-5.6 · TechCrunch coverage of the GPT-5.6 launch
1. ChatGPT WorkA new agent built into ChatGPT for longer, more involved tasks. Give it a goal and Work gathers context across your connected apps and files, breaks the job into steps, and stays with it for hours — returning finished spreadsheets, slides, documents, and even shareable web apps rather than a chat reply. You can check in, redirect it, or approve key actions as it runs.
🔗 OpenAI release notes: ChatGPT Work · Axios: OpenAI releases GPT-5.6 and ChatGPT Work tool
2. A Brand-New ChatGPT Desktop App Chat, Work, and Codex now live in a single desktop app for macOS and Windows, replacing the separate Codex app. The old desktop app still exists as "ChatGPT Classic" and keeps getting model updates and security patches, but new agent features only ship in the new app.
🔗 OpenAI release notes: the new ChatGPT desktop app
3. ChatGPT Sites (Public Beta)A new way to turn an idea into an interactive website or lightweight app without leaving ChatGPT — dashboards, project trackers, launch calendars, prototypes, and internal portals. You describe what you want, add files or constraints, preview it, then publish. It's rolling out to Pro, Pro Lite, and Edu users first, with Plus following shortly.
🔗 OpenAI release notes: ChatGPT Sites
4. GPT-Live-1 and GPT-Live-1 MiniA new generation of full-duplex voice models now powering ChatGPT Voice. Because they can listen and speak at the same time, conversations feel far less stilted — the model can interject with a quick "mhmm," handle interruptions naturally, or go quiet while you think. For anything needing web search or deep reasoning, it quietly hands off to a frontier text model in the background and folds the answer back into the conversation.
🔗 OpenAI: Introducing GPT-Live · TechCrunch: OpenAI releases new voice models
5. The Plugin Directory The App Directory has been replaced with a Plugin Directory. Plugins can now package skills, apps, and templates for specific workflows, available across Chat, Work, and Codex. Existing app connections aren't affected by the switch.
🔗 OpenAI release notes: Plugin Directory
6. Retirements and Sunsets With all this new surface area, OpenAI is also cleaning house: group chats are being phased out, the Atlas browser is being deprecated as its agentic capabilities move into ChatGPT and Codex directly, and older models like GPT-4.5 and o3 are being retired from ChatGPT.
🔗 OpenAI Help Center: ChatGPT release notes
1. The Model-Family Strategy Is a Real Shift Offering three tuned variants instead of one flagship signals that cost-per-task is now as important to OpenAI's pitch as raw capability.
2. ChatGPT Is Becoming an Output Engine, Not Just a Chat Window Work is designed to hand you finished deliverables — a doc, a deck, a live site — instead of a wall of text you still have to assemble yourself.
3. One App to Rule Them All Merging Chat, Work, and Codex into a single desktop app removes the friction of switching tools depending on whether you're chatting, delegating a task, or writing code.
4. Publishing Is No Longer a Separate Step Sites collapses the gap between "having ChatGPT help you build something" and "having something live on the web," at least for lightweight tools and internal pages.
5. Voice Is Being Positioned as a Real Interface OpenAI has been explicit that it sees voice, not just typing, as a future primary way people will manage complex, long-running agentic work — not simply a novelty mode.
6. Competitive Pressure Is Visible in the Details OpenAI's own benchmark comparisons directly reference rival models, underscoring how tightly matched — and how publicly contested — the frontier-model race has become this year.
This isn't a single feature drop — it's OpenAI restructuring ChatGPT around three ideas at once: cheaper, tiered intelligence; agents that finish real work instead of just answering questions; and interfaces (voice, sites, a unified desktop app) that get out of the way of that work. Expect the rollout to keep staggering by plan and region over the coming weeks, so if a feature isn't showing up in your account yet, it likely will soon.
This article reflects OpenAI's product releases as of mid-July 2026. Rollouts are staged by plan and region, so availability may vary and details may continue to evolve.