OpenAI used its annual developer conference on Tuesday to launch GPT-6.1 Sol and an always-on agent called Dots, framing the next phase of its business around capable systems that can work continuously and at lower cost. GPT-6.1 Sol is available immediately in ChatGPT Work, Codex and the API, while Dots are beginning a limited rollout to eligible Pro, Business Premium and Enterprise customers. The announcements moved the company’s product story away from a single frontier-model debut and toward a broader platform in which software can handle projects, use connected applications and return completed work for review.
GPT-6.1 Sol is positioned as the practical center of that strategy. OpenAI says the model approaches GPT-6 Astra’s performance on coding, computer use and professional tasks while charging one-fifth of Astra’s standard token prices. The company lists API pricing of $2 per million input tokens and $10 per million output tokens, with cached input priced at 10 cents. OpenAI’s own evaluations report gains over the earlier Sol model in software engineering, business workflows, scientific work and factual accuracy. Those are company benchmarks, not independent guarantees, but the combination of capability and cost could make advanced agent deployments more economical for businesses.
Dots extend that model shift into a consumer and workplace product. Each agent receives a cloud computer, can connect through plugins to thousands of applications and can continue working when the user is away. OpenAI says people can inspect progress, set app permissions and require approval for consequential actions. The company is emphasizing those controls because persistent agents create risks that ordinary chatbots do not: they hold context longer, operate across services and may take steps with real effects. Associated Press coverage also noted the tension surrounding the conference after OpenAI postponed a more advanced model because researchers raised security concerns.
Prediction traders read the product mix as bad news for an imminent Astra update. The Polymarket contract on an Astra 6.1-or-later release by October 9 traded near 6.7% in the latest snapshot, down about 25 percentage points over 24 hours. The broader Astra deadline event produced roughly $33,800 of activity during that period. That price does not judge GPT-6.1 Sol’s quality; it estimates whether a separate flagship model meeting the contract’s criteria will arrive before a specific date. Launching Sol while withholding the newest Astra version gave traders a concrete reason to extend the expected timetable.
The next signals will come from adoption and safety performance. Developers can now compare OpenAI’s cost and benchmark claims with production workloads, while early Dots users will test whether a continuously active agent can remain useful without overstepping permissions. OpenAI must also decide when the delayed Astra model is safe enough to release and how much detail to provide about the concerns that stopped it. DevDay showed the company can keep shipping around a flagship delay, but the larger strategic question remains: whether cheaper models and persistent agents can widen everyday use without making oversight harder.
