OpenAI’s DevDay 2026 announcements have different access timelines. GPT-6.1 Sol has launched as a lower-cost model. Eligible subscribers can initially use one dot, OpenAI’s new autonomous assistant, while texting it and creating additional dots are staged for later. OpenAI also introduced Decisions API, but its access timing and pricing remain unclear. Those distinctions matter more than treating all three announcements as tools everyone can use immediately.
The rollout at a glance
| Announcement | What is available or established | What remains staged or unclear |
|---|---|---|
| dots | One agent initially for eligible Pro and Business Premium subscribers; creation through a desktop app or browser | Text messaging is waitlisted, mobile creation is unavailable initially, and broader multi-agent access is planned for later |
| GPT-6.1 Sol | The model has launched, with published pricing for cached input | The announcement does not establish availability in every product or account |
| Decisions API | OpenAI has introduced a tool for structured decisions | General-access timing, eligibility and pricing remain unclear |
dots starts as a personal assistant with limits
A dot is an always-on AI agent: software intended to carry out tasks across connected services without a fresh instruction at every step. Powered by GPT-6 Astra, dots run on cloud computers and can learn a user’s preferences over time. OpenAI describes an ecosystem of more than 4,000 apps and plugins, with connections that include calendars, email, Slack and Microsoft Teams.
The intended shift is from asking an assistant for an answer to having it help manage ongoing work. A dot could check a schedule and messages, then draft a response or prepare a next action. That makes it potentially useful for personal organization and workplace coordination, but access to those services also makes permissions consequential. OpenAI describes policy controls, automatic review and user approval for sensitive organizational actions.
▲ Connected tasks with an approval checkpoint
The initial rollout is narrower than the always-on description might suggest. Eligible Pro and Business Premium subscribers receive one dot at first. They must create it in a desktop app or desktop browser; mobile creation is not supported initially. Texting a dot directly is behind a waitlisted beta. Additional agents are planned, but no broader rollout date is established.
Usage also depends on what the dot does. For the first 30 days, conversational chat with an initial dot does not reduce the subscriber’s usage quota. Background tasks executed through ChatGPT or Codex do count against usage limits, and standard usage terms apply after the introductory period. Someone evaluating dots should therefore distinguish a conversation from work the agent carries out.
The practical appeal may be greatest for people who want an integrated assistant without assembling their own agent system. For those already running configurable agents, the question is whether dots add enough convenience to outweigh the flexibility of choosing different underlying models. That is a workflow judgment, not an established advantage for either approach.
GPT-6.1 Sol is launched, with a specific cost claim
OpenAI positions GPT-6.1 Sol between its flagship GPT-6 Astra and the faster everyday GPT-6 Luna. Its stated proposition is near-Astra intelligence at one-fifth the operating price of Astra. The claim covers work such as coding with agents, computer use and professional tasks; it should not be read as a guarantee of identical results on every task.
For developers, the most concrete published figure is $0.10 per million cached input tokens. A token is a unit of text a model processes; cached input refers to previously processed material that can be reused. OpenAI lists that cached-input price as 95% below GPT-6.1 Sol’s standard input price and 50% below the cached-input price for GPT-6 Sol. The $0.10 figure applies to cached input, not to every kind of model usage.
OpenAI also lists monthly ChatGPT Pro plans at $100, $200 and $500. The $500 Pro 500 plan includes Astra UltraFast, an accelerated option for GPT-6 Astra. Those subscription prices are separate from the published GPT-6.1 Sol cached-input price and do not remove the need to check task quotas when using dots.
Decisions API targets narrower developer tasks
Decisions API addresses a different problem from a personal agent. An API is an interface that lets software request a service programmatically. Rather than producing an open-ended conversation, this one is designed to return a structured choice, such as yes or no, or one category from several options.
▲ Structured routing of incoming messages
A developer might use that pattern to sort incoming comments into promotional and nonpromotional groups, or to decide whether a post points to an external destination. Such classification tasks call for a specific output that another piece of software can act on. Because the output is only a short choice, decision-oriented processing of this kind can be faster and cheaper than asking a conversational model to write a full answer. A comparative performance figure, an access date and pricing have not been specified.
That leaves an important distinction for planning: dots aim to coordinate work across a person’s connected services, while Decisions API aims to help developers route repeated inputs through defined choices. One is entering an initial subscriber rollout; the other should not yet be treated as a generally available building block.
What to do next
If your account is eligible for dots, start with one limited task, review its connected-app permissions and approval settings, and check which actions consume quota before allowing background work. If you are choosing a model for repeated document processing, compare GPT-6.1 Sol’s published cached-input price with the input your workflow can actually reuse. If you need fast, structured classifications, define the categories now, but wait for clear Decisions API access and pricing details before depending on it.