OpenAI gathered developers and enterprises in San Francisco on September 29 for DevDay 2026, billed as the most ambitious edition to date. The agenda: more than twenty announcements, including persistent Dots agents, the GPT-6.1 Sol model, a new $500 Pro subscription, and a major overhaul of APIs and plugins.
🔑 Key takeaways
- Launch of Dots, always-on AI agents powered by GPT-6 Astra and integrated with more than 4,000 apps
- GPT-6.1 Sol, an agentic coding model priced roughly five times cheaper than Astra
- New Ultrafast option hitting 300 tokens/sec, 8x faster in Codex and 6x in the API
- New $500/month Pro tier with 25x the allowance of Plus and exclusive Astra Ultrafast access
- Agents API in public beta, Decisions API in preview, and a Marketplace with 32 partners including Adobe, Figma, and Salesforce
Dots and GPT-6 Astra: the era of persistent agents
The flagship product of the event is called Dots. These are persistent, always-on AI agents powered by GPT-6 Astra. Each Dot has its own cloud computer and connects to more than 4,000 applications through the OpenAI plugin ecosystem. Users can interact with these agents from ChatGPT, Microsoft Teams, and Slack. According to OpenAI, the first Dot is included at no extra cost in eligible plans.
Deployment began on September 29 for Pro subscribers in eligible regions, excluding the European Economic Area, Switzerland, and the United Kingdom. Business Premium and Enterprise versions will roll out later. This phased strategy reflects the growing regulatory constraints in Europe, notably under the AI Act, and the caution Altman has adopted after several recent security incidents involving OpenAI agents.
“We held back the latest Astra model for safety reasons: the risks tied to that level of capability are not yet fully understood.”
Sam Altman, CEO of OpenAI
GPT-6.1 Sol and Ultrafast: the price and speed war
GPT-6.1 Sol is positioned as a targeted evolution of GPT-6 Sol, geared toward agentic coding, computer use, and professional tasks. OpenAI simultaneously slashed pricing: $2 per million input tokens, $0.10/M cached, and $10/M output. That represents roughly one-fifth of Astra’s price, making agentic programming broadly accessible. The model is available via the API under the name gpt-6.1-sol, with Responses API support, tool calls, computer use, and MCP. It is not accessible from the standard Chat interface.
A new “Ultrafast” option generates tokens up to 8x faster in Codex and 6x faster in the API, peaking at 300 tokens per second. GPT-6 Astra Ultrafast is available immediately in the API and for Pro 500 and Enterprise plans. The GPT-6.1 Sol Ultrafast variant is expected in the coming weeks.
GPT-6 model pricing comparison
| Model | Input ($/M) | Cache ($/M) | Output ($/M) | Max speed |
|---|---|---|---|---|
| GPT-6 Astra | 10 | 0.50 | 50 | Standard |
| GPT-6 Astra Ultrafast | 10 | 0.50 | 50 | 300 tok/s |
| GPT-6.1 Sol | 2 | 0.10 | 10 | Standard |
| GPT-6.1 Sol Ultrafast | 2 | 0.10 | 10 | 300 tok/s (soon) |
Codex becomes an autonomous development agent
Codex continues its transformation into a full-stack development agent. Codex Cloud now lets developers run tasks remotely from a phone or from the cloud, with reusable dev environments accessible from any device, including a closed laptop. This portability is a game-changer for distributed teams and enables delegating long-running tasks even on the go.
The Codex CLI has been updated with a bidirectional voice command, a new /agents view to delegate and track multiple tasks in parallel, better prompt editing, and session resume. Code Review brings automatic summaries, diff exploration, and questions about potential issues directly into the review interface for GitHub pull requests and GitLab merge requests, with cloud-executable automated reviews.
Codex Security Cloud adds a major DevSecOps layer: scheduled GitHub repo scans, deduplication, discovery investigation, and automated fix preparation. The ambition is clear: turning Codex into an agent that no longer just suggests code, but diagnoses, patches, and audits entire codebases.
Developer ecosystem: Marketplace, APIs, and plugins
The new $500/month Pro subscription delivers 25x the allowance of ChatGPT Plus, exclusive access to GPT-6 Astra Ultrafast, and removes the 5-hour cap. OpenAI has also reopened Pro 200 sign-ups with distinct allowances for new, non-historical subscribers, signaling finer pricing segmentation.
The Marketplace, reserved for eligible enterprise customers, lets them consume part of their existing financial commitment toward approved partner software. The first 32 partners include Adobe, Figma, Sierra, Decagon, Hubspot, and Salesforce. This budget-redistribution mechanism brings OpenAI closer to the Salesforce/AppExchange model and could become a major new B2B acquisition channel.
The Agents API enters public beta with hosted execution, session management, persistent memory, tools, multi-agent support, computer use, orchestration, and context compaction. The Decisions API, in limited preview, leverages the Luna model to classify inputs, route queries, or pick an action from predefined responses, useful for intelligent routing in complex workflows.
On the plugin side, OpenAI has opened its creation platform: extensions can now occupy sidebar space and offer interactive panels. Sites can host plugins, allowing every workspace member to use the same app with their own data and permissions. MCP Events lets plugins trigger automations when an event occurs in a connected app. For collaboration, ChatGPT Space offers a shared workspace where collaborators and agents work together on files and project context, while Pages combines writing, research, chart creation, and image generation with multi-user and agent contributions. Collaborative Slides is expected in the coming weeks. The @ChatGPT integration in Slack and Microsoft Teams finally lets users mention ChatGPT in a channel, thread, or DM.
Privacy, performance, and governance
For enterprises concerned with data sovereignty, Private Intelligence includes ZDR (Zero Data Retention) combined with Private Safety Processing, enabling automated security reviews without any OpenAI staff accessing the underlying content. A preview of Private Inference, combining confidential inference with strict, verifiable controls, is slated for the autumn.
On the performance side, OpenAI announced a 45% reduction in time-to-first-token (TTFT) via the API, and tool calls and workflows more than 30% faster. The same day, OpenAI published “Towards safety cases for frontier AI training,” advocating structured safety documentation before any reinforcement learning training: automated and manual dataset reviews, immutable RL run transcripts, and chain-of-thought controls. The document notes these are goals to achieve, not a current implementation, a sign of transparency amid a regulatory storm.
Conclusion
DevDay 2026 marks a strategic milestone for OpenAI: mainstreaming always-on agents, democratizing pricing with GPT-6.1 Sol at $2/M tokens, and repositioning Codex as a true autonomous development agent. Sam Altman hinted at upcoming AI hardware, saying “something is worth the wait,” while describing Meta Muse as a “beautiful product” while maintaining confidence in OpenAI’s own lineup.
Against a backdrop of rising security incidents and calls to slow down frontier model development, the lab is trying to balance the race toward AGI with governance. Combined with the delayed IPO and a new funding round in preliminary discussions reported by CNBC, the event also highlights the colossal capital needs to sustain this roadmap. The next quarters will tell whether the Dots ecosystem and the Ultrafast option are enough to keep OpenAI ahead of Anthropic, Google, and Meta.
Sources
- Cryptoast — OpenAI DevDay 2026: key takeaways
- OpenAI — DevDay 2026
- OpenAI Community — DevDay 2026 announcements
- Agentpedia — OpenAI DevDay 2026 recap
- CNBC — OpenAI DevDay 2026 live updates
- AI Reiter — DevDay 2026 recap for developers
This article is for informational and educational purposes only. It does not constitute investment advice. Do your own research before making any decision.
