OpenAI released its GPT-5.6 model family to the public on Thursday and introduced ChatGPT Work, an agent that gathers context from a user's apps and files to produce documents, spreadsheets, presentations and web apps. The release ends a two-week preview the company had limited to a "small group of trusted partners" at the U.S. government's request. Chief executive Sam Altman told CNBC the flagship model is 54% more token efficient on agentic coding tasks and "as good or better" than competing models.

"Every enterprise now is thinking about spend and the value they're getting in exchange for AI, and this is what we really want to do," Altman said in the interview.

Key Takeaways

AI-generated summary, reviewed by an editor. More on our AI guidelines.

OpenAI prices the three tiers from $1 to $5 per million input tokens

The family ships in three tiers. Sol, the flagship, costs $5 per million input tokens and $30 per million output tokens; Terra, positioned for everyday work, runs $2.50 and $15; Luna, the smallest, costs $1 and $6. All three carry a 1 million token context window, a 128,000-token output ceiling and a February 16 knowledge cutoff. A new ultra setting runs four agents in parallel, and OpenAI's published table credits it with lifting Sol's Terminal-Bench 2.1 score from 88.8% to 91.9%.

OpenAI says Sol set a new high of 53.6 on Agents' Last Exam, an evaluation of long-running professional workflows across 55 fields, though the company's own eval table lists the score at 52.7% without labeling which reasoning configuration produced the higher figure. The same table puts Sol at 80 on the Artificial Analysis Coding Agent Index at max reasoning, 2.8 points above Anthropic's Claude Fable 5.

But Sol trails on SWE-Bench Pro, scoring 64.6% against Fable 5's 80% in the figures OpenAI itself published. Simon Willison, a developer who had early access to Sol, wrote that the model "hasn't struck me as better than Fable at the kind of complex coding tasks" he runs on Anthropic's model, and investor Matt Shumer posted on X that "for almost every task I tested, Fable was quite a bit better." Theo Browne, chief executive of the chatbot platform T3 Chat, took the opposite view, calling Sol "world-leading in computer use."

ChatGPT Work connects Slack, Salesforce and Google Drive at launch

ChatGPT Work runs on GPT-5.6 and extends Codex, the agent technology OpenAI built for programmers, to general office work. The launch post describes an agent that stays on a project for hours, breaks it into steps, and keeps working while the user is away from the computer. A plugins directory connects Google Drive, Slack, Microsoft Teams, Gmail, Salesforce and Dropbox, among others, at launch, and users can point the agent at a specific app with an "@" mention. Angela Ferrante, Zapier's head of enterprise marketing, is quoted in OpenAI's announcement saying the agent traced customer touchpoints across the company's CRM and email, found where follow-ups broke down, and "revealed seven figures in potential sales."

On web and mobile, Work reached Pro, Enterprise and Edu accounts Thursday; Plus and Business plans follow over the next few days. A rebuilt ChatGPT desktop app for Mac and Windows carries the agent on every plan, including free accounts, and folds in Codex. The prior desktop app becomes ChatGPT Classic.

Altman names Lutnick, Bessent and Cairncross in the review

The public release closes an arrangement that began in late June, when OpenAI restricted the models to government-approved organizations while the Commerce Department's Center for AI Standards and Innovation ran security testing. Altman said OpenAI worked with Commerce Secretary Howard Lutnick, Treasury Secretary Scott Bessent and National Cyber Director Sean Cairncross, describing a "collaborative back and forth" in which the government tested the models and raised problems for the company to address.

Know someone who'd find this useful? ✉️ Email it to a friend in one click, or they can subscribe free here.

A White House spokesperson, responding to an Axios report Wednesday that the administration had given OpenAI a green light, told Gizmodo that the administration "did NOT give OpenAI a 'green light,' approval, or clearance to release its models," and that "decisions on timing and scope of releases rest entirely with the companies."

Atlas browser set for August 9 deprecation

The desktop consolidation retires ChatGPT Atlas, the standalone browser OpenAI shipped for macOS last October. James Sun, an OpenAI product staff member, put the targeted deprecation date at August 9, roughly nine and a half months after the browser's debut, and said details will follow in-app and by email.

Anthropic introduced Claude Cowork, a competing agent for multi-step office tasks, months earlier, and Microsoft followed with Copilot Cowork. Meta released its Muse Spark 1.1 model on Thursday, and SpaceX shipped Grok 4.5 on Wednesday. OpenAI, which private investors value at $852 billion and which has confidentially filed IPO paperwork with regulators, began rolling out the rebuilt ChatGPT on Thursday and expects to complete that rollout within 24 hours.

Frequently Asked Questions

What is GPT-5.6?

OpenAI's model family released July 9, 2026, in three tiers: Sol, the flagship at $5 per million input tokens and $30 per million output tokens; Terra for everyday work at $2.50 and $15; and Luna, the fastest and cheapest, at $1 and $6. All three carry a 1 million token context window and a February 16, 2026 knowledge cutoff.

What is ChatGPT Work?

An agent inside ChatGPT, built on Codex technology and powered by GPT-5.6, that gathers context from connected apps and files to produce documents, spreadsheets, presentations and web apps. It reached Pro, Enterprise and Edu plans on web and mobile first; the rebuilt desktop app carries it on every plan, including free accounts.

Why was GPT-5.6's release delayed?

OpenAI limited the initial release to a small group of trusted partners at the U.S. government's request while the Commerce Department's Center for AI Standards and Innovation ran security testing. A White House spokesperson said the administration did not grant, and does not require, approval for the public release.

What happens to ChatGPT Atlas?

OpenAI is discontinuing its standalone browser, with a targeted deprecation date of August 9, roughly nine and a half months after its macOS debut. The new ChatGPT desktop app includes a built-in browser, and the prior desktop app is renamed ChatGPT Classic.

AI-generated summary, reviewed by an editor. More on our AI guidelines.

Commerce Department Clears OpenAI's GPT-5.6 for Broad Public Release
The U.S. Department of Commerce has cleared OpenAI's GPT-5.6 model family for broad public release, a person familiar with the matter told Axios on Tuesday. OpenAI announced late Tuesday night that th
The Implicator
OpenAI Restricts GPT-5.6 Release to Government-Approved Partners
OpenAI said Friday it is restricting access to GPT-5.6, its newest and most capable family of artificial-intelligence models, to a small group of partners approved by the Trump administration. The com
The Implicator
Commerce Lifts the Ban and Anthropic's Fable 5 Returns Today
San Francisco | Wednesday, July 1, 2026 Washington blinked. Eighteen days after ordering Anthropic's two most capable models offline over a jailbreak, the Commerce Department withdrew the controls,
The Implicator
AI News

San Francisco

Editor-in-Chief and founder of Implicator.ai. Former ARD correspondent and senior broadcast journalist with 10+ years covering tech. Writes daily briefings on policy and market developments. Based in San Francisco. E-mail: editor@implicator.ai