Enterprise

OpenAI ships GPT-5.6 and ChatGPT Work after Washington lifts limited-preview hold

The three-model Sol/Terra/Luna family and a new agentic workspace tool go generally available on July 9, with OpenAI pitching 'performance per dollar' as its enterprise wedge against Anthropic.

By Asher Glick · Enterprise software desk · 2026-07-17 · 3 min read

Photo: Wikimedia Commons / Steve Jurvetson, CC BY 2.0 — OpenAI CEO Sam Altman speaking at TED in April 2025

OpenAI released its GPT-5.6 model family and ChatGPT Work agent to the general public on July 9, ending a two-week limited preview that had restricted the systems to a small group of trusted partners at the U.S. government’s request. Sam Altman described the pause as producing “many changes” after a “collaborative back and forth” with the Trump administration. A White House spokesperson pushed back on any suggestion of formal sign-off, insisting the administration didn’t give OpenAI a “green light, approval or clearance,” and that release decisions “rest entirely with the companies.”

The distancing is the point. A June executive order asked frontier developers to voluntarily submit cutting-edge systems for federal capability assessments, giving agencies 60 days to stand up an evaluation process. What arrived on the other side of that window isn’t a stamped model, but a commercial rollout structured to look like one.

The new lineup is a tiering exercise: Sol as the flagship, Terra as the balanced middle, Luna optimized for speed. Sol prices at $5 per million input tokens and $30 per million output tokens, and OpenAI is pitching the whole family on performance per dollar rather than benchmark maximalism. Altman told CNBC that Sol is 54% more token-efficient than its predecessor on agentic coding tasks. The company claims Sol scores 80 on the Artificial Analysis Coding Agent Index, 2.8 points above Anthropic’s Fable 5, while using less than half the output tokens, running in less than half the time, and costing about one-third less. Sol also posts a 53.6 on Agents’ Last Exam, a benchmark for long-running professional workflows.

That framing has a specific enterprise audience. Neil Shah, VP at Counterpoint Research, told InfoWorld that rising token consumption has created “bill shocks” for enterprises, forcing them to adopt different models for different workloads. The Sol/Terra/Luna split is the answer catalog.

ChatGPT Work is the second half of the pitch. Bloomberg describes it as an agent that builds documents, spreadsheets, presentations, and web applications by gathering context from apps, files, and workflows; InfoWorld frames it as a tool that operates across applications, executes long-running tasks, and coordinates multiple tools so employees can delegate whole workflows instead of prompting step by step. It’s powered by GPT-5.6 and folds in Codex capabilities. Two new reasoning modes ship alongside: max, for complex problems, and ultra, which coordinates four AI agents in parallel. Mac and Windows desktop users worldwide, including free-tier ChatGPT accounts, get immediate access.

The 2023 voluntary White House AI commitments produced photo-op governance and little else. The June executive order and its 60-day clock look like the same pattern one iteration on: a preview window that reads, from the outside, as review, and from Washington’s podium, as nothing of the kind.

Sources