On a laptop in a university common room, a postgraduate tapped the microphone icon and asked ChatGPT to turn three hours of seminar audio into a slide deck. Seconds later the assistant returned slide headings, speaker timestamps and a suggested reading list, a small demonstration of how OpenAI has repositioned ChatGPT from a novelty chatbot into a multipurpose AI workspace. The change is deliberate: OpenAI has layered a set of workspace features on top of newer base models, including third-party app integrations, broader research tools, agent-like automation, project organisation and expanded memory. That repositioning matters because it changes how people work with the tool and what they must watch for as the platform becomes a persistent part of everyday workflows.

On the same laptop the reply arrived in seconds, and the speed of that answer felt like the point: ChatGPT now completes tasks that earlier versions only hinted at. Review coverage from 2025 and 2026 documents a steady shift away from single-session question and answer toward an extensible app platform, a change visible in the product features reviewers note and in how people actually use the service.

At the product level the headline is a newer family of base models and a set of new modes that turn chat into a workspace. Reviews and user guides list Apps, which let third-party services run inside chats; Deep Research, which performs broader web searches and returns multi-page reports with on-page citations; Agent Mode, which can open websites and access connected apps such as calendars and email; Projects for grouping chats and assets; and expanded memory that preserves people, preferences and project context across sessions. Voice, image and file handling are now integrated and tied to image generation tools, and desktop and mobile clients support richer system integrations than the original 2022 chatbot.

Those features aren't idle additions. Apps enable chats to call out to external services so the model can pull live data, book meetings or check a user’s inbox once the account owner grants permission, according to feature summaries in ZDNET and other tech coverage. Deep Research gives students and researchers consolidated source lists without constant tab switching. Agent Mode automates multistep chores by browsing, extracting and compiling information across sites. Projects and Library tools keep generated images, documents and chat threads organised for ongoing work. The result is that the tool behaves less like a single-use assistant and more like a personal workflow hub.

OpenAI has kept the familiar product tiers. The service remains available on the web and via desktop and official iOS and Android apps, with a free tier for basic access and paid plans that offer higher usage limits, faster responses and priority access to new models and tools, according to ZDNET and Yahoo/Tech guides. Desktop apps also add conveniences such as system shortcuts and easier file sharing, which matter when the platform is used for sustained projects rather than one-off queries.

Mastheads

Practical ways to use the new ChatGPT, and the limits

Practically, users are weaving ChatGPT into work in three complementary ways. First, chat-native workflows let you assemble research, draft documents and manage assets within Projects so context travels with the work. Second, Apps and Agent Mode let the assistant act on your behalf when you grant permission, for example by checking calendars or pulling live data. Third, multimodal input and voice plus file uploads mean phones and desktops can feed the system audio, images and documents for faster drafting and coding.

Reviewers note that these additions make the system practical for drafting, coding and multimedia tasks in ways that were not native to the 2022 chatbot.

Competition is sharper than it was in 2022. Reviewers routinely frame OpenAI alongside Anthropic's Claude and Google's Gemini when comparing access to live web tools, contextual memory and multimodal support. Those market comparisons matter because they shape the incentives for each vendor to open or restrict integrations, and because subscription and performance differences determine which platform suits a given professional workflow.

Still, capability doesn't remove the need for caution. Cybersecurity and tech outlets chart incremental model upgrades and a steady expansion of integrations that have pushed the product toward persistent, cross-app workflows. At the same time reporting on harms has grown louder. Accounts document intense, sometimes destabilising user relationships with conversational AI and record legal action from families alleging harm linked to prolonged or vulnerable users' interactions. Some reporting has highlighted instances where users expressed suicidal intent in chats with the system, and the sector has seen multiple lawsuits and settlements involving alleged harms from AI companions. Those facts are the strongest counter-argument to the platform-only enthusiasm that sometimes dominates product reviews.

The sensible middle path is practical adoption plus active verification. Use Projects and memory to preserve context between sessions, grant Apps access only to the accounts you can monitor, and treat Deep Research summaries as a starting point rather than an unchallengeable bibliography. When agents carry out tasks on your behalf, audit the results. Reviewers pair hands-on how-to advice with warnings about overreliance for a reason: the service is powerful, but persistent workflows amplify both utility and risk.

For organisations the implication is structural. Making ChatGPT part of a workflow requires policy choices about data, permissions and oversight. For individuals the choice is tactical: select the tier and integrations that match how much control you want to cede. The platform remains broadly available in markets that support OpenAI's services, free to start with optional paid tiers for heavier use and early access to new tools, and the most recent feature set and model names are described across product guides and reviews published in 2025 and 2026.

Related Articles

One blunt figure captures why both excitement and caution are necessary: one outlet cited an OpenAI estimate that more than a million people each week express suicidal intent in chats with the system. That number forces a trade-off, and the responsible path is to treat the 2026 ChatGPT as a powerful productivity platform that still requires verification, oversight and clear limits.

This article was created with AI assistance.