OpenAI Launches GPT-6.1 Sol at 1/5 the Price of Astra

OpenAI used this Tuesday's DevDay (29) in San Francisco to unveil two game-changing offerings for those purchasing AI at scale: the GPT-6.1 Sol, an intermediate model available through the API at $2 per million input tokens and $10 per million output tokens, and the dots, agents that remain always on, each with their own cloud computer. The company claims the event hosted over 20 announcements across ChatGPT, Codex, and various models.

Sol Nearly at Astra's Level, at One-Fifth the Price

The price of 6.1 Sol equates to one-fifth of the standard pricing for GPT-6 Astra, the company's premium model, priced at $10 and $50 per million tokens, according to The Next Web and AI Weekly. OpenAI asserts that the new Sol closely matches Astra in agent programming, computing use, and professional work; this comparison comes directly from the company and has not undergone independent evaluations. The cache input price has dropped an additional 50%, to $0.10 per million tokens, reflecting a 95% discount over the standard entry rate.

The model debuted the same day in the Plus, Pro, Business, Enterprise, and Edu plans of ChatGPT, alongside Codex, and in the API identified as gpt-6.1-sol. An Ultrafast option for Sol, up to eight times faster in Codex, has been promised for the coming days. For those running high-volume code automation, the fivefold price difference per token between Sol and Astra now outweighs the performance difference that OpenAI itself admits exists.

Dots: The Always-On Agent

The dots operate on GPT-6 Astra, possess their own browser, and connect to over 4,000 applications via plugins, according to the announcement. They handle voice calls, respond in Slack and Teams, and manage multiple projects simultaneously. The first dot is included at no additional cost in ChatGPT Pro and Business Premium.

The pricing structure has changed accordingly. OpenAI has introduced the Pro 500, a $500 per month plan with the highest usage cap available, and the only Pro plan with access to Ultrafast mode. Concurrently, the $200 Pro plan had its usage quota halved, as reported by The Next Web and Neowin. In practice, heavy users paying $200 now receive half the volume or must migrate to a tier that is 2.5 times more expensive.

Areas Where Dots Are Not Available

For now, dots will not be accessible to Pro subscribers in the European Economic Area, Switzerland, and the UK, according to a statement from OpenAI sent to the press. The company has not elaborated on the exclusion. The practical effect comes into play in any consultancy with teams on both sides of the Atlantic: a partner in New York can delegate inbox triage and client follow-up to a dot, while a colleague in Frankfurt or London is left without the tool, facing the obligations of the AI Act for general-use models effective from August 2025.

The outlook shifts in India and Brazil. Delivery centers from TCS, Infosys, and Wipro are precisely selling back-office hours to American clients, and an agent handling calls and operating 4,000 applications for $500 monthly competes for that kind of task. In Brazil, BPO operations and shared service centers catering to headquarters in the U.S. are exposed through the same channel: American clients compare the outsourced hour with a fixed monthly subscription.

No Rush for an IPO

On the same day, Sam Altman told CNBC that OpenAI would only go public when it could make "reliable safety claims" about its models and that it does not want to "rush into an IPO" while things appear "so fluid." According to Bloomberg, Altman stated that investors will be "patient." CFO Sarah Friar echoed this sentiment, noting that the company will enter the market when the timing is right for business.

The combination of announcements outlines a pricing division that corporate buyers need to recognize. Tokens become cheaper at the intermediate layer, while agents consuming Astra tokens are sold per seat, in closed plans with quotas. For those reselling AI implementation, margins shift from model consumption to orchestrating agents, which is precisely where OpenAI has just set up its own store.