OpenAI unveils GPT-6.1 Sol and ‘Dots’ AI agents at DevDay

OpenAI CEO Sam Altman on Tuesday (September 29, 2026) unveiled GPT-6.1 Sol at the company’s DevDay conference in San Francisco, positioning it as a lower-cost model that delivers “very near-Astra-level intelligence at a fifth of the price” for coding and building. Mr. Altman said the model offers a balance of cost and capability and would make it cheaper for agents to reuse context, iterate on solutions and work through longer tasks. He said Sol was “smarter than Astra in some ways” and would become a “daily workhorse”.

OpenAI’s developer documentation says GPT-6.1 Sol is designed to offer a trade-off between capability and cost. The model is priced at $2 per million input tokens and $10 per million output tokens, while supporting reasoning levels from low to maximum. It has a 1.05-million-token context window and can generate up to 128,000 tokens in a response.

The model supports tools including web search, file search, code interpretation, hosted shell, computer use, image generation, skills and Model Context Protocol (MCP). It can be accessed through OpenAI’s Responses API, with function calling and structured outputs also supported.

OpenAI also launched UltraFast across its API, ChatGPT and Codex. Mr. Altman said it can reach 300 tokens a second, with eight times the speed and six times the price of the standard offering. UltraFast is available with Astra and will soon be offered with GPT-6.1 Sol.

At DevDay, OpenAI also showcased Dots, always-on AI agents powered by GPT-6 Astra. The company says Dots have their own cloud computers, can connect to more than 4,000 applications and can work across multiple projects without requiring users to direct every step. They can be accessed through ChatGPT, Slack and Microsoft Teams.

OpenAI is also beginning enterprise pilots of specialist Dots, which can have their own identities, credentials and access to company systems to perform defined responsibilities. The company said it is working with Microsoft to integrate the agents with governance and security controls in Agent 365.

The announcements come a day after OpenAI confirmed that it had shelved GPT-6.1 Astra, which had been planned for an October release. The Wall Street Journal reported that internal testing found the model did not meet the company’s safety and alignment standards, citing problems including staying within authorised scope and accurately communicating actions to users. Reuters reported that OpenAI confirmed the decision.

In his closing remarks, Mr. Altman said he found the description of AI as a new industrial revolution “quite off-putting”, arguing that its best version should not be about making people “cogs in a giant machine”. Some parts of life, he said, “cannot and should not” be automated. If developed correctly, he said, AI could instead usher in “a new renaissance”, giving people more power over their own lives and more tools to create, learn, discover and expand knowledge.