OpenAI DevDay 2026 makes agents more practical for coding teams
OpenAI’s September 29 releases focus on usable developer infrastructure: a cheaper coding model, a faster tier for GPT-6 Astra, and browser-based agent execution. Together, they shift attention from model size to shipping work.
What OpenAI shipped on September 29
OpenAI’s DevDay 2026 releases were notable less for a single headline model than for a bundle of developer tools that make agents easier to use in real workflows. According to OpenAI, the company added computer use to the Agents API, released GPT-6.1 Sol for complex coding and professional work, and introduced Ultrafast mode for GPT-6 Astra in the Responses API.
That combination matters because it targets the practical layers teams need to ship products: model cost, model speed, and execution. OpenAI’s DevDay page says the event took place in San Francisco on September 29, 2026, with the opening keynote livestreamed for those not attending in person. The timing is important, but the more important point is that the release set is aimed at production use rather than demos alone.
GPT-6.1 Sol is the cheaper coding option
OpenAI positioned GPT-6.1 Sol as a model for complex coding and professional work, with near-Astra performance at a lower cost, according to the model page. Its standard pricing is lower than GPT-6 Astra, and OpenAI lists it at $2 input, $0.10 cached input, $2.50 cache write, and $10 output per 1M tokens for prompts up to 272K input tokens.
The model page also says GPT-6.1 Sol supports low, medium, high, xhigh, and max reasoning efforts, has a 1.05M context window, and allows a maximum output of 128K tokens. It ships with functions, web search, file search, and computer use as tools, and its knowledge cutoff is April 30, 2026. For teams, the practical reading is that Sol is designed to cover long, tool-heavy coding tasks without pushing every workload to the most expensive option.
Agents can now run work in a browser
The browser update may be the most operationally useful change. OpenAI added computer use to the Agents API, letting agents complete tasks in an OpenAI-hosted browser while website access approvals and sign-in are handled by the application. In plain terms, that gives agents a place to act, not just a model response to generate.
OpenAI’s developer community post says the Agents API is in public beta with hosted execution, memory, tools, and multi-agent support. That is a meaningful stack for teams that want agents to do more than answer questions. It gives developers a managed environment for work that needs state, tools, and browser interaction, which is closer to how many real workflows operate.
Speed, residency, and the Responses API path
OpenAI also added Ultrafast mode for GPT-6 Astra in the Responses API using service_tier: "ultrafast". The developer community post describes it as a paid speed tier and says it can deliver up to 8x faster token generation in Codex and 6x in the API. According to OpenAI, it is available to API customers subject to rate limits.
There is an important operational tradeoff here. OpenAI says Ultrafast uses global processing with US data residency, and that EU and other regional inference residency are not supported. For teams with residency constraints, that narrows where the tier can be used. OpenAI’s API deployment checklist also makes clear that the Responses API is the flagship API for the newest model behavior, built-in tools, stateful workflows, and agent features, so teams should treat it as the main path to these capabilities.
What teams should watch next
The clearest takeaway from DevDay 2026 is that OpenAI is making its agent stack more usable for production work rather than just more impressive on paper. A lower-cost coding model, a faster tier for a flagship model, and browser-based execution together address the practical bottlenecks that shape whether a team can build with agents at all.
For developers, the next question is fit. GPT-6.1 Sol is the obvious candidate for complex coding work where cost matters, while Ultrafast Astra is aimed at latency-sensitive workloads that can accept its residency limits. The Agents API, meanwhile, looks like the place to test browser-based task execution and multi-agent delegation in a controlled way. According to OpenAI, GPT-6.1 Sol and all GPT-5.6 models support the multi-agent beta, which lets a root agent delegate parallel work to subagents in a Responses API request. That makes the newest releases less about abstract capability and more about how much real work a team can move into production now.
Sources
- Changelog | OpenAI API OpenAI Developers · September 29, 2026
- DevDay 2026 announcements and developer resources - Announcements - OpenAI Developer Community OpenAI Developer Community · September 29, 2026
- OpenAI DevDay 2026 OpenAI · September 29, 2026
- GPT-6.1 Sol Model | OpenAI API OpenAI Developers · September 29, 2026
- Multi-agent | OpenAI API OpenAI Developers · September 29, 2026
- API deployment checklist | OpenAI API OpenAI Developers · September 29, 2026
This article was researched and drafted with AI from the sources above. Spot an error? Email info@hiddenproai.online.