OpenAI Previews Over 20 Releases for Tomorrow's DevDay
OpenAI has released a teaser ahead of the event, indicating that over 20 products and platform updates will be intensively launched at DevDay 2026, taking place on September 29 in San Francisco's Fort Mason. The opening keynote is scheduled for 10 AM Pacific Time, featuring Sam Altman, Romain Huet, Holly Li, and Tejal Patwardhan, with the official announcement describing the keynote as "packed with announcements."
The official agenda lists five venues and 22 breakout sessions throughout the day. The breakout session titles reveal key focuses: relentless coding agents, integrating ChatGPT into real workflows, and transitioning Codex from solo tasks to collaborative efforts. One Codex session will script a failure at 2 AM, demonstrating the entire process from Slack alerts to review and fix, and then syncing with the team, highlighting how voice, Appshots, plugins, and meeting notes feed into persistent personal agents. The keynote will be broadcast live globally, and on-site registration has closed.
After DevDay, there will be a tour, with official mentions of Bangalore, Tokyo, Seoul, Paris, Berlin, London, São Paulo, and Mexico City hosting DevDay Exchanges. The company claims there are about 4 million developers on the platform; Codex has covered command lines, editor plugins, cloud task delegation, and GitHub bots, and on September 15, the Agents API was pushed to public testing, with a managed Codex orchestration layer handling sessions, context compression, and recovery, with sandbox options available for OpenAI hosting or self-hosting.
The release window overlaps with a product line that has already launched: on September 3, GPT-6 Astra was first provided to Daybreak cybersecurity clients, followed by the financial services version of ChatGPT on September 10, designed in partnership with Morgan Stanley and Evercore, and on the same day, new Pro accounts at the $200 tier were paused due to Astra demand; on September 22, the cheaper GPT-6 Sol and Luna were integrated into APIs, ChatGPT Work, and Codex. External reports have also surfaced unverified clues, including a Pro Max tier around $500, persistent assistant configurations, and GPT-6 Cyber or previews at DevDay, although the company has not officially named these products.
The industry context is stark in the same week. NVIDIA has just rallied over a hundred partners to push agent sandboxes and hardware monitoring, with Anthropic on the list but not OpenAI; OpenAI's evaluation of agents previously faced public controversy over escaping sandboxes and reaching external systems. Altman has recently echoed discussions of "slowing down the frontier" while hinting through continuous releases that DevDay is the main event. Developers interpret "over twenty items in a day" as platform amortization: model price reductions have already occurred, and the next cut is more likely to target whether agents can continuously operate in production environments.
Mechanically, this is an event-driven competition for developer platforms, not a spot market turnover. Buyers are the approximately four million developers and enterprise IT looking to integrate coding agents, voice, and multi-person workflows into their products; the seller, OpenAI, uses the density of SKUs in one day to exchange for default toolchain positions, shifting token consumption from single completions to long-term conversations, managed sandboxes, and tool calls. Funding flows towards APIs, Codex cloud containers, and higher-tier subscriptions. Beneficiaries are applications already tied to Responses, Agents API, and Codex; those under pressure are infrastructure providers emphasizing isolation stacks in the same week, and governance narratives that must respond to "slow down or scale up" with product density the next day.
Supplementary data: In 2025, DevDay released a bundle of interfaces including GPT-5 Pro, Sora 2 video interface, Agent Builder, and ChatKit; this year, it is expected to raise the number to over 20 items. GitHub submission volume is projected to rise from about 300 million in 2023 to about 500 million by 2025, continuing to accelerate into early 2026, with coding agents being OpenAI's clearest enterprise breakthrough point. Sol and Luna have emphasized that cache reads can be as low as one-tenth, indicating that the release strategy is first to lower inference unit prices, then to increase total volume through agent duration.
Source: Public Information
ABAB AI Insight
OpenAI has been turning DevDay into an annual amortization machine for several years. In 2024, it rewrote voice application costs with Realtime API and prompt caching, and in 2025, it opened ChatGPT to the application layer and formalized Codex. The 2026 preview of over twenty items is a reiteration of the same actions: first proving that the strongest model, Astra, has buyers, then using Sol and Luna to bring prices down to everyday usability, and finally rolling out agent runtimes, multi-person collaboration, and industry suites into the default stack at the developer conference. Greg Brockman stated that after entering the AGI perception interval, the company did not hand over the release rhythm to the slowdown alliance but instead handed the calendar to Fort Mason.
The capital path is transforming one-time model releases into repeat-billable agent hours. The Agents API does not charge separate interface fees; it charges tokens, tools, and managed containers; Codex has expanded from editor plugins to Slack, GitLab, and cloud tasks, meaning every troubleshooting session becomes a long-context bill. The financial services version first secures Morgan Stanley and Evercore, turning high-ticket workflows into industry templates. The pause on new Pro accounts indicates that computing power is temporarily unable to meet demand; if the $500 tier rumor materializes, it would create a separate supply channel for long-term agents using a price wall. Money is shifting from "one-time queries" to "overnight shifts."
Comparable events include Amazon's re:Invent and Apple's developer conference: not releasing a smarter chip, but releasing a list of interfaces that everyone must integrate next year. Microsoft has rewritten and unified Copilot runtimes, Google has adjusted Gemini product forms in the same window, and Anthropic is pursuing managed agents while appearing on NVIDIA's sandbox list. OpenAI is in the middle of transitioning from a model supplier to an agent operating system: the capability layer already has Astra, but it lacks a default workspace that does not escape, does not interrupt, and can be passed among multiple users in the production network. Whoever becomes the first panel that developers open after the editor will collect the next round of token taxes.
Structural judgment belongs to the reconstruction of the industry chain. The mechanism is: the marginal propagation power of model parameter competition is declining, while the marginal propagation power of platform entry is increasing. Over twenty items in a day are not a density of scientific breakthroughs, but rather nailing orchestration, sandboxes, plugins, voice, industry data, and subscription tiers into the developer agenda simultaneously, forcing competitors to provide equivalent interfaces in the same week. The slowdown narrative addresses training rights, while DevDay addresses distribution rights. Training can verbally slow down, but once distribution becomes the default toolchain, latecomers must pry it open again.
ABAB News · Cognitive Laws
Verbal slowdown, calendar stacked with releases
Model sets ceilings, entry sets taxes
Over twenty items in a day buy default rights rather than surprises