The group of developers who received access to OpenAI’s latest model this week was deliberately small: about twenty partners, selected for the kinds of products they build, from coding assistants to enterprise automation tools. They were given a preview of GPT-5.6 Sol, a model OpenAI says will be publicly deployed within weeks, and they were asked to test it against real workloads rather than benchmark suites.
The rollout is part of a crowded week for OpenAI. The company also confirmed that ChatGPT will arrive on GenAI.mil, the U.S. military’s AI platform, in early July, the first time a mainstream commercial assistant has been formally deployed for defense personnel. And alongside the model preview, OpenAI is preparing to sell a physical product, the Codex Micro keyboard it designed with Work Louder, which goes on sale July 15.
OpenAI has not published detailed technical specifications for GPT-5.6 Sol, and the preview is governed by a confidentiality agreement that limits what partners can say publicly. The company said the model builds on the reasoning architecture introduced in earlier versions, with improvements aimed at reducing the errors that agents make when they operate autonomously, the main obstacle to wider adoption of agentic software.
The model itself is a response to the direction the industry is moving. GPT-5.6 Sol is built around reasoning and agentic use, the ability to break a task into steps, use tools and keep working until the job is done, rather than simply generating text. OpenAI has been pushing this shift for more than a year, and the new model is designed to make agents cheaper, faster and more reliable.
The partner preview is a familiar pattern. OpenAI has used staged rollouts for its previous models, giving a small group of customers early access in exchange for feedback and for real-world testing that the company’s own teams cannot replicate. The twenty partners include companies building coding tools, customer service systems and internal automation, people familiar with the program said, and their results will shape how the public version is tuned.
The public deployment is expected within July, OpenAI said, without giving a specific date. The company has learned to be careful with dates after several high-profile delays, and the phrasing, “within July,” leaves room for adjustment. The model’s name, Sol, continues OpenAI’s shift away from simple version numbers toward names that suggest a new generation of capability.
The Codex Micro hardware is the most tangible product OpenAI has shipped. The keyboard, developed with Work Louder, puts Codex commands on physical buttons, and the July 15 sale date makes it the company’s first consumer hardware product with a firm schedule. The device is aimed at developers, but its release shows OpenAI’s willingness to move into physical goods as a way of strengthening its ecosystem.
The GenAI.mil deployment is the move with the widest implications. The U.S. military has been cautious about commercial AI models, citing data security concerns that have kept OpenAI and its rivals at arm’s length for years. ChatGPT’s arrival on the platform in early July, confirmed by OpenAI and reported by Nextgov and FCW, marks the first formal use of a mainstream commercial assistant by defense personnel, and it signals how quickly the military’s posture toward AI has changed.
The military has been using AI in limited ways for years, for logistics, intelligence analysis and administrative work, but mostly through bespoke systems built by defense contractors. The decision to put a commercial assistant on GenAI.mil reflects both the capability gap, commercial models are simply better than most custom systems, and the organizational shift that has made the military more willing to buy commercial technology.
For OpenAI, the defense business is a strategic front. The company has hired defense-focused sales staff, pursued security clearances and worked with the military on pilot programs, all while managing the internal debate about whether AI companies should sell to armed forces. The GenAI.mil deployment puts that debate to rest in practice: OpenAI is a defense supplier now.
The combination of events, a new model, a hardware product and a military deployment, shows how far OpenAI has moved from its origins as a nonprofit research lab. The company that was founded to ensure that artificial intelligence benefits all of humanity now ships keyboards, sells to the Pentagon and prices its models by the token. The defense work has also raised questions OpenAI will have to answer as it expands. GenAI.mil is operated by the Defense Department, and putting ChatGPT on the platform means the company’s models will handle data from military users, with the associated requirements for classification handling and security review. OpenAI has said it will deploy the assistant within the military’s existing safeguards, but the arrangement is new enough that both sides are still defining the boundaries of what the assistant can be used for.
The twenty partners testing GPT-5.6 Sol are testing the future of the company as much as the future of the model.
