OpenAI held DevDay 2026 yesterday in San Francisco, and for once the announcements weren’t just incremental API updates. The company shipped Dots, a system for running AI agents that stay on and work while you’re not watching them. They launched Codex environments, which are actual sandboxed development machines that AI can use. And they released GPT-6.1 Sol, a new model that claims near-Astra intelligence at one-fifth the cost.
Dots is the interesting one. These are always-on agents that can monitor things, run scheduled tasks, and actually do work asynchronously. That’s different from every other AI assistant you’ve used, which forgets everything the moment you close the tab. A Dot can watch your GitHub notifications, summarize your daily code reviews, or track when a deployment finishes. OpenAI is positioning this as “AI that works for you in the background,” which sounds like marketing speak until you realize they’re actually running persistent processes on their infrastructure.
The implementation details matter here. Dots can trigger on schedules, webhooks, or events. They have access to the same tool ecosystem as ChatGPT, which means they can read email, hit APIs, and interact with third-party services. The obvious question is how much this costs when you’ve got agents running 24/7, but OpenAI hasn’t published pricing yet beyond saying it’ll be usage-based.
The other developer-focused launch is Codex environments. These aren’t just code execution sandboxes. They’re full Linux containers that persist across sessions. You can install dependencies, run databases, set up dev servers, and the environment stays alive. That means an AI coding assistant can actually test the changes it makes instead of just hoping the code works.
This fixes one of the more annoying things about AI code generation: the models have been writing code blind. They can’t run it, can’t see the errors, can’t iterate. With persistent environments, a coding agent can write a function, run the tests, see what breaks, and fix it. That’s a fundamentally different workflow from generating code snippets.
The environments integrate with the API, so you can build tools that spin up a sandbox, make changes, run a full test suite, and report back. OpenAI says these are already being used for automated code review bots that actually execute the PR changes before commenting.
GPT-6.1 Sol is OpenAI’s play for developers who want strong performance without Astra pricing. The company claims it delivers “near-Astra intelligence” at one-fifth of Astra’s token costs. That’s aggressive positioning, and it’ll come down to benchmarks and real-world testing to see if it holds up.
The model is specifically marketed for coding, computer use, and professional work. OpenAI hasn’t published detailed benchmark comparisons yet, so we don’t know exactly where it sits between GPT-6 and Astra on tasks like HumanEval or SWE-bench. But if the pricing holds and the quality is close, this becomes the default choice for high-volume API use.
The timing makes sense. Anthropic has been competitive on pricing with the Claude 3.5 Sonnet, and other providers are pushing hard on cost-per-token. OpenAI needed something between their standard GPT-6 models and the expensive Astra tier. Sol fills that gap.
OpenAI announced over 20 things at DevDay, because apparently they can’t do a focused launch anymore. Other notable items: improved ChatGPT features, new security tooling for API users, and updates to GPT-6 Astra itself. The full list is available in OpenAI’s DevDay recap.
The developer ecosystem is starting to look more like an actual platform instead of just an API endpoint. Always-on agents and persistent environments are infrastructure plays, not just model improvements. Whether developers actually want OpenAI running their background jobs and dev sandboxes instead of managing that themselves is an open question. But at least now they’ve got the option.
One email at dawn. The five stories that mattered, with the bits removed and the meaning kept. Free, for now.