OpenAI’s ChatGPT Work Takes On Claude Cowork and the SaaS Stack

OpenAI launched ChatGPT Work on Thursday, an AI agent inside ChatGPT that can gather information across connected apps and finish entire deliverables, from spreadsheets to slides to interactive sites, on its own. The agent runs on the freshly released GPT-5.6 model family and is built on top of OpenAI’s Codex technology, the same coding agent that crossed 5 million weekly users earlier this year. The release lands as a direct response to Anthropic’s Claude Cowork, the enterprise agent that kicked off the so-called SaaSpocalypse panic on India’s IT firms back in January.

The two products now compete in the same enterprise market, with the Codex coding agent folded into the new ChatGPT desktop client. The push adds a second vendor to an agent race that already cost Indian IT firms roughly $285 billion in market value during the February panic.

What ChatGPT Work Is

OpenAI’s pitch for the new agent is that it can break a large task into smaller steps, work across the web, desktop, and mobile, use templates and reference files, and keep running when the user steps away. Scheduled Tasks let it operate in the background, reviewing new Slack messages, refreshing recurring meeting agendas, monitoring customer feedback, and updating presentations when new email feedback arrives. The full ChatGPT Work launch announcement lists the kinds of work it is built for:

• Turning a discovery sales conversation into a tailored proof of concept
• Reducing month-end close and forecasting cycles
• Converting customer research into a campaign brief, then into marketing assets adapted for different markets
• Reconciling finance data across Excel and Google Sheets and turning the result into slides
• Building live dashboards, project trackers, and launch calendars through Sites in ChatGPT, a new public beta

OpenAI says nearly 100% of its own teams, including finance and sales, are already using ChatGPT Work and Codex every week. In sales, the agent turned a discovery call into a tailored proof of concept for a mission-critical problem within 24 hours, work that would normally take weeks. In finance, month-end close and forecasting dropped from days to hours after the team adopted it. Codex itself has crossed 5 million weekly users, and more than 1 million of them now use the agent for work outside software development.

Both ChatGPT Work and Codex live on OpenAI’s enterprise governance model, with admins able to set workspace-level spend controls in the Admin Console to cap usage as adoption grows. The agent carries over the same usage-based pricing as Codex, so a longer-running task consumes more of the plan’s included usage than a short chat request would. The new feature set covers the surface area Claude Cowork first claimed in January, and adds a few areas Cowork has not yet touched.

Why OpenAI Needed GPT-5.6 First

ChatGPT Work runs on GPT-5.6, OpenAI’s newest model family to reach general availability following a limited preview, per the full GPT-5.6 launch post. The family ships in three tiers: Sol, the new flagship, Terra, the balanced everyday model, and Luna, the most cost-efficient option. OpenAI’s launch announcement frames Sol as the most powerful of the three and Luna as the most cost-effective. Even at medium reasoning effort, the company says, Sol already beats Anthropic’s Claude Fable 5 by 11.4 points on the Agents’ Last Exam evaluation at roughly one-quarter the estimated cost. The pricing that makes those comparisons work is the headline number enterprises will look at first.

OpenAI’s launch post lays out the case for GPT-5.6 against the closest competition. The company’s preferred comparison is Anthropic’s Claude Fable 5, the same model family that has been the spine of the Claude Cowork launch. The numbers are not subtle:

• Agents’ Last Exam (long-running professional workflows across 55 fields): 53.6 for Sol, beating Claude Fable 5 (adaptive reasoning) by 13.1 points
• Artificial Analysis Coding Agent Index: 80 with max reasoning, 2.8 points above Fable 5
• BrowseComp (agentic browsing): 92.2%, a new state of the art
• OSWorld 2.0: 62.6%, surpassing Opus 4.8 while using 85% fewer output tokens
• SEC-Bench Pro (proof-of-concept generation on complex software): 71.2% versus GPT-5.5’s 45.8%

OpenAI also reports internal use growth that runs in the same direction. Over the past six months, the share of research compute devoted to internal coding inference grew 100-fold, while internal agentic token usage increased approximately 22-fold. Sol is being used internally to post-train other OpenAI models.

The Agents’ Last Exam is OpenAI’s preferred frame for professional work, a 55-field evaluation of long-running workflows. The benchmarks above all measure GPT-5.6 against Anthropic’s flagship reasoning model, the same model family that has been the spine of the Claude Cowork launch. On the Artificial Analysis Intelligence Index, Sol with max reasoning comes within one point of Fable 5 while completing tasks in 61% less time at roughly half the estimated cost. Ultra, OpenAI’s highest-capability setting, coordinates four agents in parallel by default to push the time-to-result down further on demanding tasks.

OpenAI vs Anthropic, Round Two

With Claude Cowork, Anthropic has been the more visible agent brand in the enterprise for most of 2026. The original Claude Cowork launched January 16, 2026, and Anthropic added 11 open-source workflow plugins on January 30. Constellation Research, covering Thursday’s launch, said OpenAI is now “looking to close the gap” with Anthropic by embedding Codex into ChatGPT and pushing the agent across web, mobile, and desktop at once. The two products are not identical.

ChatGPT Work ships on web, mobile, and desktop from launch, and the desktop tier is free. Claude Cowork started as a desktop tool for Max subscribers in January and only this week began its web and mobile rollout, per Anthropic’s Claude Cowork on web and mobile. ChatGPT Work also bundles a built-in browser and a Computer Use feature that clicks, types, and moves files locally; Atlas, OpenAI’s standalone browser, is being sunset as a result.

The fight is now a two-vendor market. Enterprises that were running pilots on a single agent product can now compare two frontier models, with Anthropic’s Sonnet 5, Fable 5, and Mythos 5 family on one side and OpenAI’s Sol, Terra, and Luna on the other. For IT services firms that spent the spring rebuilding deal structures around a one-vendor threat, that framing is the part that changes. Constellation said the move builds toward a “superapp” by combining chat, code, and work in one client. The enterprise AI engineer race is also accelerating, with OpenAI, Microsoft, AWS, and Anthropic spending $9B+ to embed engineers inside enterprise customers.

The SaaSpocalypse Returns

That positioning matters because of what Claude Cowork did to India’s IT services sector in January and February. After the 11 plugins shipped, the Nifty IT index posted its sharpest decline since the pandemic years on February 3 and 4, with Infosys, TCS, Wipro, HCLTech, LTIMindtree, and Coforge all registering meaningful single-day losses. Close to Rs 2 lakh crore in market value was erased, per Forbes India, and roughly $285 billion was wiped from global tech market capitalisation, per Entrepreneur India. Investment bank Jefferies coined the term “SaaSpocalypse” for the moment.

AI is a powerful utility, foundational infrastructure, not a finished product that can simply replace industry software. A decade ago, many believed cloud infrastructure would move up the stack and replace enterprise applications. What actually happened was the opposite: cloud became the foundation, and SaaS exploded on top of it, because compute and storage don’t equal workflows, adoption, or business results.

That was Bhanu Chopra, founder and managing director of RateGain, an AI-powered SaaS solutions provider, in Forbes India. The Indian IT response has been to restructure deals rather than stand pat. Gartner VP Analyst DD Mishra told Forbes India there is “mounting evidence that organizations are moving away from traditional multi-year transformation deals towards shorter-term, pilot-led contracts spanning three to twelve months.” On mega-deals, the emerging playbook now favors hybrid structures with modular engagements and embedded AI co-pilots.

Cognizant CEO Ravi Kumar, per Entrepreneur India, has argued the value of AI tools is still sitting with infrastructure rather than drifting into enterprises. Mphasis CEO Nitin Rakesh, in Forbes India, framed the moment as opportunity, pointing to the chance to reimagine entire business models with AI. Mastek Global CTO Ritwik Batabyal added that AI is also creating new demand in AI engineering, data, cloud, and applied domain skills.

OpenAI joining the same agent market changes the timeline for that restructuring push. Enterprises that were negotiating shorter pilot-led contracts against a single vendor now have two competing agent platforms to compare. The shift from time-and-effort billing to outcome-linked models, long the fear in the services sector, gains a second vendor behind it. The next round of mega-deal renewals, due through the end of 2026, will be the first test of whether the SaaSpocalypse thesis holds up against a more crowded field.

What Costs $5 per Million Input Tokens

ChatGPT Work connects to Slack, Microsoft Teams, Google Drive, SharePoint, email, calendars, CRMs, and project trackers through a unified plugins directory. Users can type “@” plus the app name to direct the agent to a specific source, and a feature called Auto-review uses OpenAI’s most advanced models to vet important actions involving connected tools and APIs before they happen. The pricing on the launch webcast, per Constellation Research:

The standalone Codex app is being merged into the new ChatGPT desktop client. The old desktop app will be renamed ChatGPT Classic. The Codex coding experience, including inline editing within diffs, pull request review in the side panel, faster computer use on GPT-5.6, and support for multiple repositories in a single project, stays intact for developers. A new Chrome extension brings ChatGPT into the browser’s sidebar, and Sites in ChatGPT, the new public beta for turning work or ideas into interactive web apps you can share via URL, completes the productivity push.

OpenAI’s chat help center says the new desktop experience is now available globally for Mac and Windows. ChatGPT Work on desktop is open to every plan, including Free; the web and mobile versions stay on the paid tiers. Sites in ChatGPT and the upgraded browser are the two most visible new pieces, and both target the same kind of cross-app workflows Claude Cowork was built for.

What Rolls Out When

The schedule, per the OpenAI launch post. ChatGPT Work enters the market in two waves: web and mobile on the paid tiers, desktop on every tier including the free plan.

Enterprise and Edu admins can set workspace-level spend controls in the Admin Console to cap usage. OpenAI said ChatGPT Work follows the same usage-based pricing structure as Codex, with more complex tasks consuming more of the plan’s included usage. Admins can also support high-impact work across teams without raising limits broadly by setting workspace-level defaults, configuring group limits, and creating individual overrides for people who need more capacity. The company is positioning the rollout as a phased, admin-led deployment.

Frequently Asked Questions

What is ChatGPT Work?

ChatGPT Work is an AI agent inside OpenAI’s ChatGPT that can gather information across connected apps, plan multi-step tasks, and produce finished deliverables such as spreadsheets, slides, documents, and interactive sites. It is powered by the GPT-5.6 model family and built on top of OpenAI’s Codex technology.

How is ChatGPT Work different from Claude Cowork?

ChatGPT Work ships on web, mobile, and desktop from launch, with a free desktop tier. Claude Cowork, Anthropic’s competing agent, started as a desktop tool for Max subscribers in January 2026 and only began its web and mobile rollout in July 2026. ChatGPT Work also bundles Sites, a feature for turning work into shareable interactive web apps, and a Computer Use capability that clicks and types inside local apps.

What is GPT-5.6?

GPT-5.6 is OpenAI’s newest model family to reach general availability, with three variants: Sol (the flagship), Terra (a balanced everyday model), and Luna (the most cost-efficient option). Launch pricing on the webcast was set at $5 per million input tokens for Sol, $2.50 for Terra, and $1 for Luna.

What is the SaaSpocalypse?

The “SaaSpocalypse” is a term investment bank Jefferies used to describe the market reaction to AI agents entering enterprise software workflows. After Anthropic released 11 Claude Cowork plugins in late January 2026, India’s Nifty IT index posted its sharpest decline since the pandemic years on February 3 and 4, and roughly $285 billion in global tech market value was wiped out in the panic.

Will ChatGPT Work replace IT services and SaaS?

OpenAI is positioning ChatGPT Work for cross-tool workflows that span finance, sales, marketing, and HR, with plugin connectors to enterprise apps. Indian IT firms including Cognizant, Mphasis, and Mastek have argued that the value of enterprise integration, accountability, and change management still sits with services companies, and that the shift will produce shorter pilot-led deals focused on AI integration services.