OpenAI Offers Subscription-Based Cloud Agents and Unveils GPT-6.1 Sol

OpenAI launches customizable cloud agents available 24/7 and expands its range with GPT-6.1 Sol, an intermediate evolution presented as being close to Astra at a significantly lower cost. Ultrafast versions, a dedicated Pro plan, and security safeguards accompany these new offerings, with availability varying by plan.

Costs, Speeds, and Safeguards: Sol 6.1 and Ultrafast Pricing

GPT-6.1 Sol is offered at $2 per million tokens for input and $10 per million for output, which is five times cheaper than Astra. Cached repeated segments are charged at $0.10 per million tokens, representing a 95% reduction. According to OpenAI, GPT-6.1 Sol achieves Astra-level performance on the DeepSWE programming test for about one-fifth of the cost, improves by 7 points over its predecessor in computer commands, and is 2.1 points behind Astra. For scientific research tasks, an execution costs an average of $5.47, compared to over $23 for Astra and Claude Opus 5.5, with Astra maintaining the highest score. When faced with tricky questions, GPT-6.1 Sol has a factual error rate of 4.1%, compared to 4.5% for GPT-6 Sol and 4.0% for Astra. OpenAI also offers Ultrafast versions of Astra and Sol, which are up to eight times faster, accessible via the API, ChatGPT Work, and Codex with a Pro plan at $500 per month. In terms of security, the model is classified as critical in cybersecurity and retains Astra's safeguards; for tasks that encourage dishonesty, OpenAI reports a misleading code presentation in 1.5% of cases, compared to 0.5% for Astra.

Targeted Availability: Where to Use GPT-6.1 Sol and Dots

GPT-6.1 Sol is available starting today for Plus, Pro, Business, Enterprise, and Edu subscribers, but only in ChatGPT Work and Codex, and not in the standard chat. OpenAI has not yet announced a date for its arrival in the latter. Ultrafast options are available via the API as well as in ChatGPT Work and Codex, with a Pro plan at $500 per month. Dots are being rolled out today to Pro and Business Premium subscribers in the relevant countries, and are in testing with Enterprise clients. The first dot is included in the subscription, and interactions with it do not count against the ChatGPT message quota. Pro subscribers can also sign up for a waitlist to use iMessage and RCS.

Control and Segmentation: How Dots Manage Trust

The dot's computer is separate from the user's, unless the latter chooses to connect them. For compatible services, the agent uses saved passwords without the model having access to them. When acting on its own initiative, its tools cannot send messages or modify applications. Custom rules allow for certain actions to be authorized, require approval, or block them, and operations like changing passwords remain reserved for the user. Specialized dots for businesses are entering a pilot phase, and integration into Microsoft’s Agent 365 tool is in preparation. OpenAI clarifies that these agents can make mistakes.

Uses and Channels: Virtual Colleagues Available Continuously

Powered by GPT-6 Astra, the dots operate 24/7 and have a browser as well as their own credentials to access services. They can write and test code, manage multiple projects simultaneously, and identify tasks to be completed; a tester indicated that a dot detected an unbilled publication, prepared the invoice, and then sent it after approval. Users can interact with them in ChatGPT, via SMS, on Slack and Teams, or even by call, and the context of a conversation is preserved across channels.

Positioning: Managed Cloud vs. Local Open Source

OpenClaw is mentioned as a pioneering open-source solution installed locally and compatible with various AI models. The dots, on the other hand, operate in OpenAI's cloud, rely on GPT-6 Astra, require no installation, and benefit from company-managed safeguards. They connect to over 4,000 applications and can be named and customized, including with visual elements like a beret, glasses, or large eyes.

Mapping the GPT-6 Range and Timeline

The GPT-6 range now includes Astra as the most powerful model, Luna in a lightweight version, and Sol as the intermediate option. GPT-6.1 Sol, an enhanced version of Sol, aims to get closer to Astra without a price increase, while GPT-6.1 Astra has been postponed for security reasons. OpenAI emphasizes a more affordable intelligence with this variant. The entire lineup was presented during DevDay 2026, held on Tuesday, September 29, in San Francisco, just days after Anthropic's announcement regarding Sonnet 5.5. The dots, continuous agents based on GPT-6 Astra, and the Ultrafast variants are among the main announcements.