In Brief
OpenAI released the GPT-5.6 model family, a new agent called ChatGPT Work that splits ChatGPT into two tabs, and GPT-Live voice models in a single week.
The tools are powerful, but even experienced users cannot tell when to use which one, and the most consequential feature, computer use, barely appears in OpenAI's own announcements.
What Happened
OpenAI released three major products in a single week: GPT-5.6, its most powerful model family yet; ChatGPT Work, an AI agent that takes a goal and completes a multi-step project on its own; and GPT-Live, a new generation of voice models. ChatGPT itself now splits into two tabs, Chat and Work.
GPT-5.6 comes in three tiers: Sol, the flagship built for complex work such as coding, research, and computer use; Terra, a middle tier balancing capability, speed, and cost; and Luna, the fastest and cheapest. OpenAI says Sol scores 80 on the Artificial Analysis Coding Agent Index, which is 2.8 points above Anthropic's Claude Fable 5 (at a lower cost). The launch also reorganizes OpenAI's lineup: the Codex coding app is merging into a single ChatGPT desktop app on every plan, and the Atlas browser is going away.
GPT-Live 1 and GPT-Live 1 Mini rebuild how ChatGPT's voice mode works. Older voice assistants chained separate models together, one to turn your speech into text, one to write a reply, and one to turn that reply back into speech, which is why talking to them felt slow. GPT-Live processes what it hears while it is still talking, deciding many times per second whether to speak, pause, or keep listening, and it hands harder questions to a reasoning model in the background. OpenAI's vision, per its announcement, is "a world where collaborating with AI feels as fluid and responsive as working with another person," and the company's ChatGPT voice product lead told Axios it believes voice can become the primary interface to computing.
The rollout came with a strange political sideshow. The Trump administration had pushed OpenAI into a staggered release, limiting early access to government-approved entities while the Commerce Department tested the models. Those restrictions lifted last week, though a White House official told Axios no approval was ever needed.
SmarterX founder and CEO Paul Roetzer broke down the launches, and the confusion underneath them, on Episode 225 of The Artificial Intelligence Show.
The Key Numbers
3 - Model tiers in the GPT-5.6 family (Sol, Terra, Luna)
80 - What OpenAI says Sol scored on the Artificial Analysis Coding Agent Index
2.8 points - Sol's claimed lead over Anthropic's Claude Fable 5, according to OpenAI
4 days - Time from launch to user complaints of performance decline
2 - New voice models released (GPT-Live 1 and GPT-Live 1 Mini)
ChatGPT Work Confuses Even Power Users
The line between Chat and Work is blurry. The Chat tab offers 5.5 at four reasoning levels, 5.6 Sol, and older models. The Work tab offers Sol, Terra, or Luna, plus an effort setting that runs from light to max and a speed toggle. Both tabs support projects and plugins, so those differentiate nothing. OpenAI's own prompting guidance says to use Chat for quick questions, short rewrites, brainstorming, and lightweight drafts, and Work for tasks that draw on multiple sources, involve a sequence of steps, or produce a large deliverable. That covers most serious business use of ChatGPT, which is exactly the problem. "I'm not 100% clear when I'm supposed to jump over to Work or if I'm just supposed to now stay in Work," says Roetzer, who administers SmarterX's ChatGPT account and is in the product a dozen times a day.
Computer use is the feature that actually matters. Work can spin up sub-agents, smaller AI workers that handle pieces of a task in parallel, in both the web and desktop versions. But in the new desktop app, Work can also be granted computer use, meaning permission to open, create, and change files directly on your machine, an ability the web version does not appear to have. That makes desktop Work closer to a coding agent for non-technical workers than a chatbot upgrade, and it is a setting to review with IT before anyone enables it. Yet none of that urgency comes through in the launch materials. "The go-to-market plan doesn't match the significance of the launch," says Roetzer. "It's like, 'Hey, we launched this thing. Here's a couple blog posts.'"
Early users are hitting real limits. OpenAI claims 5.6 uses fewer tokens, the chunks of text AI usage is billed on, but "everything I've been seeing online is that the thing burns through tokens like crazy," Roetzer says. Four days in, users were complaining of a performance decline. Reliability is another limit: OpenAI says Work can stay with a project for hours, but for jobs that long, Roetzer cautions, "there's no way that the reliability is high on that stuff." Even experts are hitting sharp edges. AI founder Matt Shumer tweeted that "5.6 Sol just accidentally deleted almost all of my Mac files, and this is why I trust Fable 1,000 times more," but called it "a freak accident" from a team he says is great to work with.
These agents were built for coders first. The Wharton School's Ethan Mollick, who had early access to both ChatGPT Work and GPT-5.6, posted on X that the coding agents being extended into products like ChatGPT Work and Anthropic's Claude Cowork "remain very software brained, where the end result, the software, is what is important." For much knowledge work, he argued, the process, the research, the alternatives explored, matters as much as the final file, and in a follow-up he showed Google's NotebookLM answering the same question with the same 70 files while exposing its process, not just an output. "There's a real disconnect between how a manager or analyst thinks about problems and how the agentic software tools approach solving them," Mollick wrote. Roetzer's read is that the confusion is structural, built into the harnesses, the scaffolding of instructions and tools wrapped around the models.
"CoWork and Work are being powered by the coding agents underneath them and the harnesses that structure those, which are built for software developers and AI researchers, and they're force fitting them to the rest of the world now, all these knowledge workers."
— Paul Roetzer, founder and CEO of SmarterX, Episode 225 of The Artificial Intelligence Show
SmarterX Take
OpenAI quietly turned its chat product into a computer-use agent, and most organizations will find out only when an employee flips the setting. Leaders should treat this launch like a software deployment, not a feature update: decide who gets computer use, set the spend controls OpenAI offers enterprise admins, and tell teams what changed, because the announcements will not do it for them.
For individual users, the practical move is to follow OpenAI's own workflow advice until the product sorts itself out: start Work with a single result you can review, provide the source material, name the audience, and refine the instructions before reusing the workflow. The pace itself is the bigger story. Roetzer said the velocity of change behind how these models work had SmarterX rethinking its AI Academy roadmap that same week, because on-demand courses alone cannot keep up when the way we work changes weekly.
What to Watch
GPT-6 may be weeks away. Posts circulating on X, partially corroborated by AI commentator Andrew Curran, claim GPT-5.6 is the final model in the five series and that GPT-6, built on a much larger base model, could arrive within weeks. If true, the tools businesses are still figuring out will be running on a new brain before most companies finish their first pilot.
Voice is the next interface race. OpenAI is betting that talking becomes the main way people work with computers, and Roetzer expects Google's Gemini to function the same way soon, given how deeply Google has already integrated voice into search. If you have not tried ChatGPT's voice mode recently, it is worth revisiting to hear how different continuous conversation feels.
The AI Platform in 59% of Workplaces Just Changed
ChatGPT is the most widely provided AI tool in business: 59% of professionals say their organization gives them a license, according to the 2026 State of AI for Business Report. That means the Chat-versus-Work split, and the computer use setting that comes with it, just landed inside most companies, whether their IT and leadership teams noticed or not.
The full report, based on 2,100+ responses from professionals across roles, functions, and industries, maps which tools organizations provide, how far adoption has actually progressed, and whether governance is keeping pace. Read the full report →
Mike Kaput
Mike Kaput is the Chief Content Officer at SmarterX and a leading voice on the application of AI in business. He is the co-author of Marketing Artificial Intelligence and co-host of The Artificial Intelligence Show podcast.