OpenAI DevDay 2026 Recap: Always-On Dots Agents, GPT-6.1 Sol Pricing and the Ultrafast Speed Tier

On September 29, 2026, OpenAI used DevDay 2026 to announce more than 20 major items across ChatGPT, Codex, its models and new ways for people and AI to work together. The company calls it its biggest DevDay yet. It also said it is opening ChatGPT as a shared surface where humans and agents collaborate and where developers can launch native experiences directly to a collective 1.2B weekly users. One note on sourcing: this report rests on the captured text of OpenAI's own recap page. That capture is cut off inside the Code Review section, and many details are absent, including full pricing, plugin specifications and benchmark scores. This report does not guess at them. Where we offer analysis, we label it as ours. The first item is Dots. OpenAI describes them as remarkably capable, always-on agents built to handle everything. The pitch is that a Dot gets to know what matters to you, works on your behalf all the time, and takes important work off your plate so you get time and attention back. Dots are available on Pro and Business Premium in eligible markets. Enterprise, Edu and Healthcare users can try a beta once their workspace admin enables it, and it is off by default. That default matters. An always-on agent holds context and permissions for a long time, so the first control an enterprise needs is an administrator switch, not a per-user opt-in. The second item is GPT-6.1 Sol. OpenAI calls it a major upgrade to GPT-6 Sol, with exceptionally strong performance on agentic coding, computer use and professional work. It is said to deliver near-Astra intelligence to everyone at a fifth of Astra's standard input and output token prices, so developers can run more complex workflows within the same budget. It is available to all API, Plus, Pro, Business, Enterprise and Edu users. The captured text gives no unit prices and no benchmark numbers, so the fifth is a relative claim from the vendor. Our analysis: when near-flagship capability arrives at roughly a fifth of the price, many teams will move most routine agent traffic to Sol and keep Astra for the hardest tasks. Model routing then becomes a standard design step.

The third item is Ultrafast, a premium speed tier for workloads where speed matters most. OpenAI states up to 8x faster token generation in Codex, at 300 tokens per second, and up to 6x in the API. GPT-6 Astra Ultrafast is available today in the API, and in ChatGPT Work and Codex on Pro 500 and Enterprise plans. GPT-6.1 Sol Ultrafast is coming soon. For agentic coding the benefit is easy to see: a long task is a chain of many steps, and the wait at each step adds up, so faster generation shortens the total run. The multipliers are ceilings, though, and the captured text does not state any price premium. Confirm both before you plan capacity. The fourth item is Private Intelligence, which helps businesses use frontier AI with greater confidence that their data is protected. Two parts are disclosed. The first is Zero Data Retention with Private Safety Processing, which enables automated safety reviews without giving OpenAI personnel access to the underlying content. The second is a preview of Private Inference, coming this fall, which combines confidential computing with strict, verifiable controls. The second part has a timeline but no technical detail yet. Our view: regulated sectors such as finance and healthcare will focus on the word verifiable. The real hurdle is whether auditable evidence is available, not whether a promise is made.

The fifth group covers Codex. Developers can now run Codex wherever they need it: on a computer, remotely from a phone, or in the cloud from any device. Reusable development environments help tasks start quickly and give a team a shared setup with approved settings and permissions. This is available on Plus, Pro, Business, Healthcare, Education and Enterprise. The refreshed Codex CLI lets you start and steer tasks by voice. A new /agents view makes it easier to delegate work and track several tasks at once. Everyday flows for editing prompts, resuming sessions and using worktrees are better, and a cleaner terminal UI helps in longer sessions. It is available to all plans. A new code review experience in the ChatGPT desktop app lets you read summaries, explore diffs and ask Codex about potential issues before you share feedback on GitHub pull requests or GitLab merge requests. The text about automatic reviews is cut off in the capture.

The sixth point is ecosystem strategy. OpenAI says it is expanding its commitment to an open ecosystem. ChatGPT becomes a shared surface for people and agents and a distribution channel where developers launch native experiences. Plugins add more customization inside ChatGPT, and the recap also lists more to do with a ChatGPT subscription. The site navigation still shows the Apps SDK and open models under Developers. For developers, 1.2B weekly users is a large audience. But revenue share, review rules and data ownership are not mentioned in the captured text, so wait for the official documents before you commit a roadmap. The industry effect falls into three parts. First, the product form moves from question and answer to ongoing delegation. The test for quality shifts from one good reply to long-run reliability and permission governance. Second, price and speed are now layered together: Sol takes high-volume traffic at a low price, while Astra and Ultrafast serve high-value and low-latency cases. That squeezes mid-range pricing for other vendors. Third, privacy is becoming a precondition for enterprise purchase, and Private Intelligence shows a leading vendor turning keep the data inside the trust boundary into a product. Outlook and risks: an always-on agent costs more when it errs, because it acts while nobody watches. How open the distribution channel really is depends on commercial terms not yet published. The true price-performance of Sol and Ultrafast needs independent testing. A sensible path is to pilot Dots and cloud Codex on low-risk workflows first, log every failure, and widen their permissions in steps.