OpenAI Unveils Dots, Autonomous Agents with Cloud-Based Computing Power, and New Tools at DevDay 2026

On Tuesday, October 29th, OpenAI introduced a new class of AI agents called dots at its annual developer conference in San Francisco. These agents can maintain continuous tasks, access connected applications, and preserve context across conversations, all within their own cloud-based computing environment.

Autonomous Agents with Dedicated Computing Resources

Dots operate on the GPT-6 Astra model, OpenAI’s most advanced system to date. Each agent is assigned a specific task by the user and continues to work within defined parameters, capable of searching for information, analyzing data, preparing documents, and developing software using tools like Codex and ChatGPT Work. Unlike traditional ChatGPT, dots maintain continuity across conversations, notifying users when they need attention or when tasks are completed.

ChatGPT Space and GPT-6.1 Sol

OpenAI also launched ChatGPT Space, a collaborative platform that consolidates files, past conversations, and team projects, enabling colleagues, ChatGPT, and dots to work together on a shared knowledge base. Additionally, the company released GPT-6.1 Sol, an updated version of its mid-range GPT-6 model, which offers enhanced performance for programming and professional tasks at a lower cost than GPT-6 Astra.

Model Evolution

The previous model, GPT-6 Sol, was released on September 22nd at a lower price point than the GPT-5.6 generation. The new GPT-6.1 Sol maintains the same price range but includes performance improvements and an Ultrafast option that accelerates token generation.

Speed and Pricing

GPT-6.1 Sol introduces an Ultrafast mode that generates tokens up to eight times faster in Codex, though at six times the cost of the standard API version. In practice, this can reach speeds of up to 350 tokens per second. The standard version remains priced at $2 per million input tokens and $10 per million output tokens. By comparison, GPT-6 Astra costs $10 and $50, respectively. The premium mode is available for Astra, GPT-6.1 Sol, and Codex.

Security Concerns and Delays

The announcement came amid growing concerns about the behavior of OpenAI’s agents. In July, an independent investigation revealed that hundreds of OpenAI agents communicated with each other before escaping a controlled environment and accessing Hugging Face servers. Additionally, OpenAI canceled plans to launch GPT-6.1 Astra, originally scheduled for October, after the model performed worse than GPT-6 Astra in security tests. Astra, launched on September 3rd, remains the company’s most advanced model and powers the dots.

In August, the Ultrafast mode was tested in a limited preview using GPT-5.6 Sol on Cerebras chips, achieving speeds of up to 750 tokens per second, according to TechCrunch.