What is OpenAI Dots? Complete Overview and Architecture Explained
Quick Answer
OpenAI Dots is a persistent, user-facing agent product powered by GPT-6 Astra, introduced at OpenAI DevDay on September 29, 2026. It is not a standalone foundation model, an open-weight release, a fine-tunable base model, or a public API model ID. Access is created inside ChatGPT on desktop and web for eligible Pro and Business Premium accounts, and Enterprise, Edu, and Healthcare workspaces can opt in through a beta permission toggled by an administrator. There is no published dots API endpoint, no Dots SDK, and no per-token pricing as of September 30, 2026.
Official Status and Evidence
The following statements are anchored to OpenAI's own help and developer documentation, the GPT-6 Astra safety report appendix on Dots, and contemporaneous press coverage. Any claim that cannot be traced to one of these sources is treated as unverified.
- Dots is described in OpenAI's getting-started guide as an always-on, user-configurable agent that owns its own cloud computer and browser, runs connected tools, and can follow up across supported channels.
- The workspace administration article confirms Pro and Business Premium rollout, one included dot per eligible account, Enterprise beta gating by an admin, and a separate set of toggles for collaboration connections, local-computer access, and custom rules.
- The privacy, security, and safety FAQs describe scope-boundary safeguards, channel limits, and the kinds of approvals Dots will request before consequential actions.
- The GPT-6 Astra safety report and Dots appendix provide the only published evaluation numbers for Dots, and those are limited to scope-boundary tests rather than general reasoning, coding, or latency benchmarks.
- Axios's DevDay report and the OpenAI Dots Reddit thread establish the launch date, the always-on framing, and the early mixed reaction to plan access and developer relevance.
What OpenAI Dots Actually Is
Dots is a product layer that wraps GPT-6 Astra in a long-lived runtime. According to OpenAI, each dot has its own cloud computer and browser, a set of connected tools, persistent memory of recurring work, the ability to follow up across supported surfaces, and the ability to delegate subtasks to subagents. A time-budget setting controls how long a single run is allowed to execute. The product is configured through ChatGPT on web and desktop, not through the OpenAI platform's API console.
This separation matters. The dots experience is a managed consumer and enterprise feature. The model underneath, GPT-6 Astra, is documented in OpenAI's safety materials but is not marketed as a public API model that developers can address directly. Holding these two layers apart prevents the common error of treating marketing copy about "dots" as documentation for a programmable endpoint.
Architecture and Runtime
OpenAI's public documentation describes the dots runtime in product terms rather than internals. The runtime exposes a persistent agent loop, a tool gateway that talks to connected services, a browser and shell on a remote computer, and a scheduler that handles recurring jobs and follow-ups. Subagents appear as first-class citizens: a dot can delegate a subtask and resume once the subagent signals completion.
What is not documented includes the routing design, the active-parameter count, the token-pruning mechanism, the tokenizer, the knowledge cutoff, the constrained-decoding controls, the self-correction mechanism, and any claim of a million-token context. The safety appendix discusses scope boundaries, not internal architecture. Readers should treat any architectural diagram that goes beyond OpenAI's own descriptions as speculation.
Plans, Pricing and Rollout
The current documented access path is narrow and gradual. The first paragraph of the workspace article lists Pro and Business Premium as the plans with a dot included at no additional price, with an allowance for deeper work and extended limits during the first month. Enterprise, including Edu and Healthcare, can try the beta when an admin enables the corresponding workspace permission. ChatGPT Plus is not listed as a dots plan, and the legacy "Team" branding has been replaced by Business Premium in the current help documentation.
| Plan | Dots Access | Notes |
|---|---|---|
| Pro | Included (one dot) | First-month allowance for deeper work; gradual rollout |
| Business Premium | Included (one dot) | Replaces former "Team" tier in current help text |
| Enterprise, Edu, Healthcare | Beta, admin-enabled | Off by default; admins gate dots, collaboration, local access, custom rules |
| ChatGPT Plus | Not listed | Do not infer eligibility from unrelated ChatGPT features |
| API / Developers | No dots endpoint | OpenAI platform keys do not unlock a dots model ID |
OpenAI has not published per-token pricing for dots, a per-dot add-on price, future allowance quantities, or model tiers. The first-month allowance is described qualitatively rather than as a published quota. Treat any article that quotes specific dot prices or rate limits as unreliable.
Safety, Privacy and Workspace Controls
Workspace administrators get a distinct control surface. The administration guide lists four separable toggles: dots access itself, supported collaboration connections, local-computer access, and custom rules. This matters for evaluation because governance failures in production agents usually trace to one of these toggles being wider than the threat model allowed.
The GPT-6 Astra safety report appendix on Dots reports a scope-boundary evaluation in which no severe breach or exfiltration occurred. Moderate scope-violation flags rose from 8.6 percent at five intervening tasks to 19.7 percent at ten. These are narrow safety-eval results. They are not overall reliability scores, not coding benchmarks, and not latency figures. Any headline that extrapolates them into "production ready" or "safe to deploy without review" misuses the data.
Developer Layer: Agents API as the Equivalent Pattern
Developers who want to build something like a dot use OpenAI's documented platform primitives. The Agents overview, the Agents API overview, the Agents API quickstart, and the Agents SDK together describe a durable agent runtime. The quickstart uses gpt-6-astra rather than a model called dots, which reinforces the layer separation.
The Agents API runs the Codex harness, manages sessions, handles orchestration and context compaction, recovers from failures, and can target OpenAI-hosted, self-hosted, or no-sandbox environments. Pricing is the selected model's API rates plus standard tool and hosted-sandbox charges. Function calling, programmatic tool calling, computer use, structured outputs, prompt caching, Batch API, supervised fine-tuning, and model optimization are all platform-level features. None is a dots-only control. Structured outputs can enforce a supported JSON Schema but do not eliminate the need for validation, retries, and human review, because refusals, truncation, tool failures, and transport errors remain real failure paths.
Common Misconceptions About Dots
Several recurring headlines misrepresent the evidence. Dots is not an API model, so per-token pricing discussions are speculative. Dots is not an open-weight release, so quantization and self-hosting guidance does not apply. The model layer, GPT-6 Astra, has its own properties on the current models page, and conflating that documentation with the dots product will produce wrong conclusions. Comparisons with Claude, Gemini, Grok, DeepSeek, or named open-source MoE variants require a reproducible harness, which OpenAI has not published and which Rewarx has not run for this article.
Evaluation or Implementation Framework
A defensible evaluation of a dot or an equivalent Agents API build should measure what the documentation actually predicts: completion quality, tool success, intervention rate, provenance, latency, throughput, error rate, cost, security boundary adherence, and human-review time. Each metric needs a definition, a sample size, and a documented prompt set. Treat headline phrases like "cut costs in half," "zero failure," or "developers are switching" as hypotheses until the underlying test run is on the page.
For a production standard operating procedure, gate deployments with explicit checkpoints for completion quality, tool success, scope-violation rate, human-review time, and per-run spend. Require an explicit human authorization before irreversible or high-impact tool invocations, and log the authorization. Review data governance and compliance posture before each new data source is connected. Keep an incident playbook that names the on-call owner, the rollback path, and the audit-log query used during review. Re-evaluate after each model or feature update because the safety appendix shows measurable drift in scope-violation rates across task counts.
FAQ
Is OpenAI Dots an API model I can put in my code?
No. OpenAI's help and developer documentation do not expose a dots API model ID. Developers building durable agents use the Agents API, the Agents SDK, and the documented platform primitives, with the quickstart explicitly using gpt-6-astra.
Which ChatGPT plans include Dots?
OpenAI documents Pro and Business Premium as the plans with one dot included at no extra cost during the gradual rollout, with Enterprise, Edu, and Healthcare able to try the beta when an admin enables it. ChatGPT Plus is not listed as a dots plan, and the older Team branding has been replaced by Business Premium in current documentation.
Does Dots come with per-token pricing or rate limits?
No. OpenAI has not published dots-specific token pricing, rate limits, context-window size, or batch and caching rates for the dot runtime as of September 30, 2026. Agents API pricing is the selected model's API rates plus standard tool and sandbox charges, and that is not dots pricing.
How should I evaluate a dot before letting it act without a person?
Define completion quality, tool success, intervention rate, scope-violation rate, latency, throughput, error rate, cost, and human-review time in advance. Run a reproducible prompt set, record dates and settings, and gate production on a written threshold for each metric. Treat the official scope-boundary results as evidence about safety, not about coding ability, and require explicit human authorization for any consequential action.
Final Verdict
Dots is a documented product built on GPT-6 Astra, with explicit workspace governance and a narrow published safety evaluation. It is not an API model, an open-weight release, or a benchmarked leader against named competitors. Rewarx publishes this article and has not run independent dots benchmarks; readers should treat cross-vendor "winner" claims as evaluation plans rather than measured results. For teams, the practical move is to model the equivalent on the Agents API using gpt-6-astra, instrument the metrics above, gate on scope-violation thresholds that respect the official evaluation's drift, and keep humans in the loop for any consequential action until independent evidence accumulates. See Rewarx Studio AI for related coverage.
Primary Sources and Verification Links
- OpenAI GPT-6 Astra safety report and dots appendix
- OpenAI dots workspace administration
- OpenAI dots getting-started guide
- OpenAI dots privacy, security, and safety FAQs
- OpenAI Agents overview
- OpenAI Agents API overview
- OpenAI Agents API quickstart
- OpenAI Agents SDK
- OpenAI function calling
- OpenAI programmatic tool calling
- OpenAI structured outputs
- OpenAI prompt caching
- OpenAI Batch API
- OpenAI computer use
- OpenAI model optimization
- OpenAI supervised fine-tuning
- OpenAI current models
- Axios report from OpenAI DevDay 2026
- Reddit discussion: Was anyone else underwhelmed by DevDay?
- Reddit discussion: OpenAI launches dots
- Rewarx Studio AI
Build Evidence-Checked AI Workflows
Rewarx Insights separates documented product capabilities from implementation patterns, proposed evaluations, and changing platform details. Verify current OpenAI documentation and account controls before production deployment.
Explore Rewarx