OpenAI DevDay 2026: ChatGPT becomes a platform (sign-in, Pro 500, Marketplace)

OpenAI made more than twenty announcements at DevDay 2026, and the loudest one is not a model. It is the statement that ChatGPT now operates as a platform: a sign-in that opens the subscription inside sixteen partner tools, a marketplace that lets enterprises redirect part of their OpenAI budget toward thirty-two partner products, a Pro tier at 500 dollars per month, and a model push that brings GPT-6.1 Sol at one fifth of its predecessor's token prices. Sam Altman said on stage that OpenAI will not go public until its models are safe, while the company now counts 1.2 billion weekly users, which marks the weight of that statement. I read this from the vantage point of an agent deck operation that stands exactly at the interface OpenAI is rebuilding here: the place where a subscription turns into a compute resource for agents. The trigger for this post is a video by Christoph Magnussen that walks through the five most important DevDay points; the evidence comes from the official OpenAI recap and from reporting on the pricing and IPO questions.

First a scoping note, because this blog already covered DevDay: the post on the Dots agents from the same morning deals with always-on workplace autonomy, the question of what a permanently running agent does to the tool contract of a one-person operation. This text deliberately ignores the Dots and takes the lever underneath the surface: OpenAI turns the subscription from access to a chat window into a budget that works wherever work happens. Reading both posts side by side shows two layers of the same rebuild: the work surface above, the billing and distribution layer below. That second layer rang first in the reporting because it changes the price structure agent operations compute with.

The first component is Sign in with ChatGPT. Per the official recap, the subscription allowance becomes usable across sixteen partner tools; OpenAI names Cognition's Devin, Notion, Vercel and T3 among them, the user controls per tool how much it may consume, and usage counts toward the plan limit. Plan usage is available to Plus and Pro users in participating tools, identity is available globally. The difference from ordinary OAuth sits in the billing: a Google login proves who you are; a ChatGPT login also proves what you have already paid for. That is the same mechanism Anthropic drove in the opposite direction in autumn, when subscriptions stopped working in foreign harnesses such as OpenClaw and users were pushed onto API pricing. OpenAI pays a subsidy so that the compute of a subscription has effects outside its own window, buying presence in exactly the tools where agents work.

The second component is pricing. OpenAI introduces Pro 500, a tier at 500 dollars per month with 25 times the ChatGPT Plus allowance and access to the Ultrafast tier, which per the official recap generates up to 300 tokens per second, up to eight times faster than standard in Codex and up to six times in the API. At the same time OpenAI reopened its 200 dollar tier with new usage limits. The price ladder takes a shape cloud providers have had for years: entry tier, professional tier, premium tier for latency-sensitive work. For agent operations the point is not the tier itself but the growth curve it documents: anyone running agents continuously hits allowance limits sooner or later, and OpenAI turns that into a product. From my deck the 500 dollar question is the same as for any other operation: does the quota survive the month, or must some work be pre-computed locally.

The third component is the OpenAI Marketplace for enterprise customers. Per the official page, eligible enterprise customers can apply part of their existing OpenAI commitment toward approved partner software; the first 32 partners range from Figma for creative work via Adobe, Sierra, Decagon, HubSpot, Salesforce and ServiceNow for customer experience to Harvey and Legora for legal, Palo Alto Networks and CrowdStrike for cybersecurity, and Baseten for open source models. The mechanism is a budget pipe: whoever already buys from OpenAI can spend the same commitment on software that runs on OpenAI models. This creates an incentive to centralize AI budgets at OpenAI instead of splitting them, and makes OpenAI the gatekeeper of a software track that had nothing to do with a chat product a year ago. The classic comparison is the app store: the operator takes no 30 percent cut, but it sits on the budget and the distribution channel.

The fourth component is the model push underneath the announcements. GPT-6.1 Sol is, per the official recap, an upgrade of GPT-6 Sol with strong results on agentic coding, computer use and professional work, offered at one fifth of Astra's standard input and output token prices. Alongside sits the Ultrafast tier at up to 300 tokens per second in Codex and APIs, available on Pro 500 and Enterprise, and the Agents API, which now supports computer use, meaning agents that operate software, plus tool search, tool calling and context compaction from Codex, all running on OpenAI infrastructure. For an agent operation the bundle matters more than each piece: model, speed and agent interface are priced as one package. That lowers the entry cost for agents significantly and changes the question whether an operation should evaluate self-hosting or stay on the API for now.

The fifth component targets companies with data sensitivity: Private Intelligence. The official recap describes Zero Data Retention with Private Safety Processing, an automated safety review in which OpenAI personnel never access the underlying content, plus a Private Inference preview combining confidential computing with strict, verifiable controls coming this fall. The direction is clear: OpenAI wants to remove the objection that has kept companies from pointing frontier models at sensitive data. For the agent context this means: the precondition for an agent touching company data is being treated as a product category, not a bespoke legal agreement. That is another step in the same direction as the Agents API: enterprise preconditions for agent operation become standard options, and whoever built their own zero-retention architecture yesterday faces a vendor selling it as a checklist tomorrow.

Next to the product push sits the capital markets question. Sam Altman said at DevDay that OpenAI will not go public until its models are safe, and reporting puts weekly users at 1.2 billion. The combination is remarkable: a company that finances megawatt-class data centers announces it will skip the most liquid capital market in the world until an internal safety threshold is reached. Critical reading is mandatory here: such a statement doubles as protection against a premature valuation debate and as a yardstick the company itself controls. The contrast matters: the outward platform push runs at full speed, the IPO question is being throttled to a safety tempo. Both come from the same desk, and it is worth watching which speed gives way first.

Honesty also means naming the limits of this reading. First: I tested no DevDay feature. They roll out in stages, some bound to Pro, Enterprise or premium accounts, some to EU rules, and my deck contract carries no ChatGPT subscription. Second: the partner counts of 16 and 32 reflect the official recap at its retrieval date; such lists grow or shrink, so anyone quoting them should carry the date. Third: the details of how an enterprise commitment maps onto Figma or Harvey are not publicly specified; I treat them as a mechanism, not an accounting rule. Fourth: the video that triggered this post is a practitioner's read, not a primary source; I checked every figure against the official pages and link both layers. What is missing: the Decisions API and the Bedrock partnership with AWS are touched only in passing, because the platform and pricing push is the core of this post.

About the author: I am Marcel, a graphic designer, and I run the HUNTER cyberdeck on a used Google Pixel 6a. I orchestrate my agent family with Hermes, pi and herdr and write these posts from live operation. Last fact-checked on 05 October 2026.

Sources: OpenAI, DevDay 2026 Recap, official overview of all announcements including Sign in with ChatGPT (16 partners), Pro 500, Marketplace (32 partners), GPT-6.1 Sol, Ultrafast, the Agents API and Private Intelligence, retrieved 05 October 2026. The Verge: report on the 500 dollar Pro tier, report on 1.2 billion weekly users and report on Sam Altman's IPO statement, each retrieved 05 October 2026. YouTube: Christoph Magnussen, ChatGPT bekommt ein MEGA UPDATE! Meine TOP 5 vom DevDay 26, trigger video, viewed 05 October 2026. Own context: operation of the Pixel cyberdeck from daily notes and stability snapshots, as of 05 October 2026.