Analyst(s): Nick Patience
Publication Date: July 10, 2026
OpenAI has launched the GPT-5.6 model family (Sol, Terra, Luna) alongside ChatGPT Work, a merged ChatGPT desktop app and a new Sites feature that turns a prompt into a finished spreadsheet, deck, or shareable web app. The launch reframes ChatGPT around delivered outcomes rather than answers, and arrives with benchmark and pricing claims that place GPT-5.6 directly against Anthropic’s Claude line.
What Is Covered in This Article:
- OpenAI’s launch of GPT-5.6 (Sol, Terra, Luna) and ChatGPT Work, an agent built to turn a stated goal into finished sheets, slides, docs, and web apps.
- The use cases OpenAI demoed: finance forecasting and variance analysis, launch-readiness synthesis, notes reorganization via computer use, live translation, design prototyping, and a six-month case study with a Japanese farm.
- A merged ChatGPT desktop app that folds in Codex, rebrands the prior client as ChatGPT Classic, and begins sunsetting the Atlas browser.
- How GPT-5.6’s self-reported benchmarks and pricing compare with Anthropic’s Claude Fable 5, Opus 4.8, and Mythos 5.
- What ChatGPT Work’s cloud-first, multi-surface design means competitively against Anthropic’s Claude Cowork, Microsoft 365 Copilot, and Google Workspace.
The News: OpenAI launched ChatGPT Work, a new agent inside ChatGPT built to convert a stated goal into finished output, spreadsheets, slides, documents, and interactive web apps, rather than a conversational reply. The launch coincided with the general availability of GPT-5.6, OpenAI’s newest model family, offered in three tiers: Sol, the flagship model; Terra, positioned for everyday work; and Luna, the fastest and lowest-cost option. ChatGPT Work runs on GPT-5.6 and incorporates Codex, OpenAI’s coding agent, letting it gather context across connected apps, local files, and browser tabs, then work independently for extended stretches while pausing for approval on key decisions.
Alongside ChatGPT Work, OpenAI introduced a merged ChatGPT desktop app for macOS and Windows that brings Chat, Work, and Codex into one interface, renamed its prior desktop client ChatGPT Classic, and began sunsetting its standalone Atlas browser. A companion feature called Sites enables users to turn data and analysis into shareable, interactive dashboards, prototypes, and web apps. The rollout follows a roughly two-week delay tied to the U.S. government’s review of GPT-5.6 and is arriving in stages across ChatGPT’s Pro, Enterprise, Edu, Business, and Plus tiers, with web and mobile access following the initial desktop release.
OpenAI ChatGPT Work Ships Files, Not Just Chat. The Enterprise Race Is On
Analyst Take: ChatGPT Work is OpenAI’s attempt to reposition ChatGPT away from chat and toward delivered output. That’s more important than another round of benchmark one-upmanship, because it’s a bet that enterprise buyers will pay for finished artifacts, not just better answers. The demo lineup was built to sell that proposition; nearly every segment ended with a spreadsheet, a slide deck, or a shareable site, rather than just text.
The Use Cases Are the Pitch
OpenAI’s own demos leaned hard on finance: a variance analysis reconciled across systems, an updated Excel forecast, a PowerPoint, and a matching Sites dashboard, all from one request. Recruiters were shown using scheduled tasks to chase interview feedback, and data scientists using on-the-fly visualization for quick requests. On the desktop app, ChatGPT Work synthesized a large ticketing export into an interactive visualization, then, in a separate example, pulled a launch-readiness PDF, user-research interviews, a security review, and three open Chrome tabs into a slide deck that matched the presenter’s own template in about 90 seconds, reportedly drawing on memory from earlier conversations to fold in context that had never been shared via Slack. A design team member showed ChatGPT Work turning Slack and Gmail context into an interactive site in a single prompt, then live-annotating it into a playable, mini 3D scene, a livelier update to the long-running community test of asking a model to draw a pelican riding a tricycle.
OpenAI’s additional customer materials added examples from Virgin Atlantic saying it used ChatGPT Work to benchmark customer journeys against competitors, compressing a multi-week research cycle into hours; Zapier built a lead-triage system that traces prospects across its CRM, email, and call-recording tools into a weekly executive dashboard; and an NVIDIA events lead automated registration and post-event feedback analysis for GTC, freeing up time previously spent on manual spreadsheet work. Outside the office entirely, OpenAI featured a Hokkaido broccoli farm that has used ChatGPT for six months to coordinate fieldwork and team communication, plus an on-stage live interpretation between English and Japanese, a reminder that OpenAI is pitching ChatGPT Work as a general-purpose agent, not solely a knowledge-worker tool. Internally, OpenAI said Sol autonomously kicked off and ran a post-training job for Luna from a short, underspecified prompt, and that Codex generated the plots and slide deck used in the livestream itself, alongside cybersecurity results from OpenAI’s Project Daybreak and Patch the Planet initiatives, which give outside researchers early model access and channel the results into accepted patches for open-source projects, including Linux.
Does It Change Competitive Math?
At the model layer, GPT-5.6’s self-reported benchmarks put Sol roughly level with Anthropic’s Claude Mythos 5 on coding-style evaluations, while Luna reportedly outperforms Claude Opus 4.8, though it trails OpenAI’s own prior flagship on the same test. Independent reaction has been mixed rather than decisive: some early testers reported by Axios rate Anthropic’s Fable 5 as the more capable model on raw reasoning, while others find GPT-5.6 the more dependable choice for everyday production work. None of this settles which model is better so much as it confirms that OpenAI and Anthropic are trading benchmark leads within a point or two of each other, release after release, which puts more competitive weight on the product wrapped around the model.
That is where ChatGPT Work’s architecture is a genuine point of differentiation. ChatGPT Work is cloud-hosted and multi-surface by design, reachable from web, mobile, and desktop, with a built-in browser and connector access aimed at producing shareable artifacts, sheets, slides, docs, and now Sites. Anthropic’s answer, Claude Cowork, takes a different path: it runs locally inside the Claude desktop app with direct file-system access and a plugin model built on Claude Skills, an architecture that appeals to data-sensitive verticals such as legal and financial services that are wary of routing sensitive files through a cloud agent. Neither approach is straightforwardly superior; the choice trades cloud reach and shareability against local control and data residency. It’s the kind of trade-off enterprise buyers, not benchmark tables, will end up deciding.
Futurum’s ETR data adds urgency to OpenAI’s timing. ETR’s AI Product Series data showed OpenAI’s models remaining the most widely used foundation model in enterprises as of March 2026, but with Anthropic’s enterprise usage share climbing from 21% to 48% in a year, the fastest gain of any vendor tracked. ChatGPT Work is arriving as OpenAI’s usage lead is already narrowing, which raises the stakes of converting a near-billion weekly user base into stickier, workflow-embedded habits before Anthropic’s Cowork and Claude Code momentum closes further ground.
Against Microsoft and Google, the pressure runs in a different direction. By generating finished Office-format spreadsheets, decks, and now standalone web apps directly inside ChatGPT, OpenAI is now competing on ground Microsoft has historically owned through native Word, Excel, and PowerPoint integration. Microsoft’s previous dependency on OpenAI’s models is now a three-way hedge between OpenAI, Claude, and its own MAI models. If ChatGPT Work can produce finished artifacts without the Microsoft 365 wrapper, the case for paying for both gets harder for some workflows, though Copilot’s governed, in-document integration, complete with track changes and Microsoft Graph context, remains a real advantage ChatGPT Work does not yet replicate. Google’s Gemini keeps its own edge in long-context research and native Workspace grounding, an area ChatGPT Work’s connectors and Sites feature don’t directly address.
What Doesn’t Change
GPT-5.6’s rollout was delayed roughly two weeks pending U.S. government review, in the same window that a separate export-control action briefly took Anthropic’s Mythos 5 offline. Frontier launches now seemingly need to clear a US government checkpoint, regardless of vendor. ChatGPT Work’s expanded action surface, browser access, local files, and third-party connectors also raise the same governance questions enterprise buyers have been asking of Anthropic’s own agent rollout: auditability, artifact review, and connector-permission scoping will matter more to procurement teams than another benchmark chart. And the economics are still settling: OpenAI shifted Workspace Agent runs to token-based credit pricing on July 6, just three days before this launch, so how efficient ChatGPT Work actually feels to a paying enterprise will depend on real-world token consumption once usage moves past staged and free-preview access.
Whether ChatGPT Work moves the needle, in the end, depends less on GPT-5.6’s benchmark position, which is already converging with Claude’s, than on whether OpenAI can turn its user base into workflow-embedded, connector-governed enterprise habits faster than Anthropic can extend Cowork and Claude Code, or Microsoft and Google can deepen their own native productivity-suite advantages and their respective ecosystems will be an important part of that too.
What to Watch:
- Whether GPT-5.6 access completes its staged rollout to Plus, Business, and free tiers on schedule, and whether admin controls keep pace with ChatGPT Work’s expanded connector and file access.
- Independent verification of OpenAI’s self-reported efficiency and coding benchmarks once Sol, Terra, and Luna are broadly available through the API.
- How Anthropic responds: whether Claude Cowork, Claude Code, or its next model release adjusts pricing or scope in reaction to ChatGPT Work’s cloud and mobile reach.
- Whether Microsoft repositions Copilot’s native Office integration as a governance advantage now that ChatGPT Work generates finished Office-format artifacts directly.
- The August 9, 2026 shutdown of the standalone ChatGPT Atlas browser, and whether its agentic-browsing capabilities transfer cleanly into ChatGPT Work.
- Continued government review of frontier model releases, following GPT-5.6’s delay and the brief export-control action against Anthropic’s Mythos 5 in late June.
See the complete announcement on ChatGPT Work on the OpenAI website.
Disclosure: Futurum is a research and advisory firm that engages or has engaged in research, analysis, and advisory services with many technology companies, including those mentioned in this article. The author does not hold any equity positions with any company mentioned in this article.
Analysis and opinions expressed herein are specific to the analyst individually and data and other information that might have been provided for validation, not those of Futurum as a whole.
Other Insights From Futurum:
How Anthropic and OpenAI Are Building “Everywhere Ecosystems”
Anthropic’s Stainless Acquisition Signals a New Phase in Agentic AI Competition
Anthropic Files For IPO, Looking to Beat OpenAI to the Punch
Author Information
Nick Patience is VP and Practice Lead for AI Platforms at The Futurum Group. Nick is a thought leader on AI development, deployment, and adoption - an area he has researched for 25 years. Before Futurum, Nick was a Managing Analyst with S&P Global Market Intelligence, responsible for 451 Research’s coverage of Data, AI, Analytics, Information Security, and Risk. Nick became part of S&P Global through its 2019 acquisition of 451 Research, a pioneering analyst firm that Nick co-founded in 1999. He is a sought-after speaker and advisor, known for his expertise in the drivers of AI adoption, industry use cases, and the infrastructure behind its development and deployment. Nick also spent three years as a product marketing lead at Recommind (now part of OpenText), a machine learning-driven eDiscovery software company. Nick is based in London.
