OpenAI Sets DevDay 2026 Keynote as Teaser Points to 20-Plus Launches

OpenAI is entering DevDay 2026 with a teaser for more than 20 launches, putting the focus on whether the company will turn its latest model, coding and agent releases into a broader developer platform. The company has scheduled the event for Tuesday, September 29 in San Francisco, with a 10 a.m. PDT keynote featuring Sam Altman and free livestream access. OpenAI has not published the launch list, so the 20-plus slate should be read as a pre-event signal rather than confirmed product detail. The official event page says DevDay will bring developers together to test new work, attend technical sessions, and meet the teams building OpenAI tools; in-person applications are closed, invited attendees were quoted $650 registration, and the company says DevDay Exchanges will follow in Bengaluru, Tokyo, Seoul, Paris, Berlin, London, Sao Paulo, and Mexico City. The context makes the teaser unusually important. Over the past two weeks, OpenAI has already shipped major building blocks that look like runway for the event: GPT-6 Sol and GPT-6 Luna cut API prices by 50% from GPT-5.6 promotional pricing, with GPT-6 Sol priced at $2 per million input tokens and $10 per million output tokens and GPT-6 Luna at $0.10 input and $0.50 output. OpenAI says GPT-6 Sol scored 33.2% on AutomationBench at $0.27 per task, beat Claude Opus 5 at max effort at 9% of its cost per task, reached 56.4% on Agents' Last Exam at 60% lower cost than Claude Opus 5, and scored 68.8% on DeepSWE v1.1 at about 80% lower cost than Claude Fable 5's top score. The models are available in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise and Edu users, while Free and Go users can access GPT-6 Luna in the desktop app; OpenAI says they are not yet available in Chat. The platform direction is just as significant as the models. The Agents API gives applications access to an OpenAI-managed Codex harness that manages sessions, orchestration, context compaction and recovery, while agents can work in sandboxes, run code, edit files, connect to MCP servers and produce artifacts. It supports subagents, steering while work is underway, and session resumption, and it bills model usage, OpenAI tools and OpenAI-hosted sandboxes separately. That framework explains why developers are watching for products that move beyond one-off chat into durable workflows. The most closely watched unannounced item is o, described in unreleased ChatGPT configuration references as an always-on assistant with o as a display name and -o as an email suffix, but OpenAI has not confirmed its tasks, permissions, memory model, pricing or launch scope. Other possible DevDay threads remain unconfirmed, including new Platform onboarding tiers, broader Ultrafast API availability and a reported Pro Max plan that would sit above current ChatGPT subscriptions. Reception among builders is split between excitement about cheaper inference and caution about practical limits. Lower token prices, 90% cached-input discounts and GitHub's reported reduction of fresh prompt processing by more than 50% across billions of requests are being treated as useful for agents that reuse large contexts. At the same time, developers are asking whether usage allowances, context windows, permissions, auditability and production reliability will match the promise of longer-running agents. The pressure is sharpened by OpenAI's own Pro tier notice: new subscriptions and upgrades to the $200 Pro 20x plan have been paused since September 10, while Pro $100 remains available with 5x Plus usage and Pro $200 remains the highest usage tier for existing users. DevDay's central question is therefore not simply how many launches OpenAI shows, but whether it can package models, agents, sandboxes, pricing and controls into a coherent platform that developers can trust for real work.