OpenAI DevDay 2026: What's Actually New for Developers
October 1, 2026
OpenAI used its September 29 DevDay to ship more than 20 announcements at once, but only a handful actually change what you can build today. The clearest wins for developers are GPT-6.1 Sol, a near-frontier model at a fifth of the price, and computer use landing in the Agents API OpenAI had opened in public beta three weeks earlier. Codex also picked up a cloud runtime, a refreshed CLI, and a dedicated security-scanning product.
Key Takeaways
- GPT-6.1 Sol replaces GPT-6 Sol (released just a week earlier) with near-Astra performance on agentic coding and computer use at roughly a fifth of Astra's token price, available now as gpt-6.1-sol
- The Agents API, already in public beta since September 10, gained computer use, letting agents operate software through its UI, plus multi-agent delegation, tool search, and automatic context compaction
- Codex now runs in the cloud from any device with reusable dev environments, alongside a refreshed CLI (voice control, an /agents view) and a new Codex Security Cloud for scheduled repo scanning and automated fixes
- Decisions API, a new option for real-time classification and routing against a fixed set of answers using the Luna model, is in limited preview
- Bedrock Managed Agents lets you run OpenAI's agent stack natively inside AWS, and Ultrafast (up to 8x faster in Codex, 6x in the API) is live today for GPT-6 Astra
What's new for the API and models?
GPT-6.1 Sol is the headline model release, and it's a fast follow: GPT-6 Sol and Luna only shipped a week earlier, on September 22. OpenAI says Sol now delivers performance close to GPT-6 Astra on agentic coding, computer use, and general professional work, at roughly a fifth of Astra's standard input and output token prices. It's live immediately under the model ID gpt-6.1-sol in the API, and also available in ChatGPT Work and Codex across Plus, Pro, Business, Enterprise, and Edu plans. Alongside it, Ultrafast — a premium speed tier promising up to 8x faster generation in Codex (around 300 tokens per second) and 6x faster in the API — is live today for GPT-6 Astra, with a Sol version coming soon.
What changed in the Agents API?
The Agents API itself isn't new — OpenAI put it into public beta on September 10, giving developers the same harness and infrastructure that runs Codex, without having to build session management, orchestration, or recovery themselves. What DevDay added is computer use: agents built on the API can now interact with software through its interface rather than only through predefined tool calls. That sits alongside multi-agent delegation, tool search (which loads tool definitions on demand to save tokens instead of keeping every definition in context), and automatic context compaction that trims earlier turns as a session nears its context limit, so long-running agents don't need custom summarization logic. You pay for the tokens and tools your agents use; there's no separate API fee. Computer use specifically is available through the API and in Codex and ChatGPT Work on Pro 500 and Enterprise plans.
What's new in Codex?
Codex picked up three separate updates. Codex Cloud lets you run a task from a computer, a phone, or the cloud, using reusable development environments so a team can share an approved setup instead of everyone configuring their own. The Codex CLI got a refresh: voice control for starting and steering tasks, a new /agents view for tracking multiple tasks at once, easier prompt editing and session resumption, better worktree support, and a cleaner terminal UI for long sessions. Separately, Codex Security Cloud gives defenders a way to scan entire GitHub repositories on demand or on a schedule, with Codex investigating findings, deduplicating them, and preparing fixes in the cloud — including access to Daybreak Blue models without needing a separate Daybreak application.
What else did OpenAI ship for developers?
Decisions API is a narrower tool than the Agents API: instead of open-ended generation, it uses the Luna model to pick an answer from a fixed set you define, aimed at classification, routing, or choosing an agent's next action from text or image input. It's in limited preview today with wider access coming soon. On the infrastructure side, Bedrock Managed Agents — built with Amazon — lets you run OpenAI's agent stack natively inside AWS, combining the Agents API's core capabilities with AWS-specific customization and integrations. OpenAI also opened up plugin extensions, letting developers give a ChatGPT plugin its own sidebar entry and interactive panels, and added support for the proposed MCP Events specification so plugins can trigger automations from events in connected apps.
What about Dots and the rest?
Most of the remaining announcements — Dots (OpenAI's always-on personal agent), ChatGPT Space, Pages, collaborative slides, team tasks, and @ChatGPT in Slack and Microsoft Teams — are consumer and workplace-collaboration features rather than things you build against. They're worth knowing about if you're evaluating ChatGPT as a product for your team, but they don't change what's available in the API or Codex.
Wrapping Up
If you only have time to try one thing from DevDay, it's probably GPT-6.1 Sol — swap it in wherever you were running Astra for agentic tasks and see if the price drop holds up on your own workload before committing further. Computer use in the Agents API is the one to watch next, since it's the clearest sign OpenAI wants the harness behind Codex to become infrastructure everyone builds on.
Sources
Frequently Asked Questions
What is GPT-6.1 Sol?
A September 29, 2026 upgrade to GPT-6 Sol with performance close to GPT-6 Astra on agentic coding, computer use, and professional work, at about a fifth of Astra's token price. It's available now as gpt-6.1-sol in the API.
Is the Agents API new?
No — it launched in public beta on September 10, 2026. DevDay added computer use support to it, along with multi-agent delegation, tool search, and automatic context compaction.
What does computer use in the Agents API let an agent do?
It lets an agent interact with software through its actual interface — clicking, typing, navigating — rather than being limited to predefined tool calls.
What is the Decisions API for?
Real-time classification and routing. Instead of open-ended text generation, it uses the Luna model to pick one answer from a fixed set you define, based on text or image input. It's in limited preview.
How is Codex Cloud different from the Codex CLI?
Codex Cloud is a hosted runtime so tasks can run from a computer, phone, or the cloud with a shared, reusable environment. The Codex CLI is the refreshed terminal tool for steering tasks locally, now with voice control and an /agents view.
What is Bedrock Managed Agents?
A joint OpenAI-Amazon product that runs OpenAI's agent stack natively inside AWS, adding AWS-specific customization and integrations on top of the Agents API's core capabilities.