OpenAI DevDay 2026: Dots, GPT-6.1 Sol, and Ultrafast Announced

OpenAI combines personal agents, new models, and distribution tools during its DevDay. Dots is now available for Pro and Enterprise customers, GPT-6.1 Sol is launched, and the Ultrafast mode promises up to 300 tokens per second. Marketplace, API Decisions, Sites, and Codex security complete a set of releases available today or in the process of deployment.

Security and Deployment: 53 Fixes and a Codex Update

OpenAI reports having conducted a security sprint a few months ago involving a quarter of its product engineers, with 53 critical issues fixed on the first day. This initiative, called The Defense Factory, has led to a major update of Codex Security Cloud announced during DevDay. The new version offers scheduled scans, automatic deduplication, and comes with an open-source Codex Security CLI tool. OpenAI also indicates that it is building a security knowledge base that integrates architectural details not visible in the source code and uses SECURITY.md files to define threat models and guidelines by product. A security scan was conducted live to illustrate these features.

Distribution: Marketplace, ChatGPT Connection, and Expanded Sites

OpenAI is launching a Marketplace with partners and reactivating the connection with ChatGPT, allowing users to connect to third-party applications using their existing tokens. The publisher also offers plugin extensions that act as native applications for ChatGPT and Codex, with Adobe cited as an example. ChatGPT Sites, available since the beginning of the year, already hosts 8 million sites, and 70% of OpenAI employees are developing their own Sites. The Sites offer the ability to convert a spreadsheet into an application, be continuously updated by Codex, have an SQLite database, and scheduled tasks launched during DevDay. The connection with ChatGPT is supported, Sites are private by default, and plugins for Sites are launched to display user-specific data. It is possible to add editors to publish updates.

Models and Performance: Launch of GPT-6.1 Sol and Ultrafast Mode

OpenAI announces GPT-6.1 Sol, touted as offering intelligence close to Astra for a fifth of the price, and available today. The Ultrafast mode is launched, reaching up to 300 tokens per second, advertised as eight times faster and priced six times the standard rate. Ultrafast is accessible via the API, ChatGPT, and Codex, available for Astra 6 immediately and for Sol 6.1 soon. An overview of the API Decisions is also unveiled: by providing the Luna model with a predefined set of options, it aims for responses in a fraction of a second. These announcements are part of a platform strategy covering large models, tools, and distribution.

Personal Agents and Collaboration: Dots and ChatGPT Space

OpenAI introduces Dots, customizable agents represented by an avatar, powered by the Astra model. A demonstration included voice interactions and an API migration case. Sam Altman describes Astra as OpenAI's most aligned model and asserts that one can entrust their Dot with as many responsibilities as desired. For teamwork, ChatGPT Space serves as a shared environment, featuring a slash-type menu. A demonstration led by Holly Li showcased a Dot named Dottie with access to ChatGPT, Slack, and Teams, capable of using Codex on a laptop to build an application and launch it in the iPhone simulator. Dots deployed in Slack have their own identities. Access to Dots is open to Pro and Enterprise customers. OpenAI is working on specialized Dots for legal and finance, plans collaborations with companies, and is conducting a project around Microsoft 365.

Developer Tools and Research: Open Source Codex and API Agents

OpenAI is making Codex open source and specifies that the tool is now fully cloud-based. An API for Agents integrating computing usage is available. Regarding research acceleration, OpenAI states it is using its own models: Tejal Patwardhan cites optimization of the computing harness, achieving a latency reduction by a factor of 2, and a decrease in navigation errors on desktop and browser. Privacy features aimed at enterprises have also been mentioned. Sam Altman presents OpenAI's API as the most performant and reliable at the moment. On the offerings side, a Pro 500 subscription grants access to Ultrafast with, according to OpenAI, 25 times the usage of the Plus plan, while the $200 per month plan is being resold for new subscribers.

Demonstrations and Metrics: 1.2 Billion Weekly Users

Romain Huet conducted a series of demonstrations, including a 3D modeling scene powered by Ultrafast, despite a non-functional voice mode. Commands allowed for displaying a livestream in a 3D model of the room, and an application distributing free tickets was coded live. OpenAI showcased the computing use of Astra to discover a new game, the construction of an iPhone Duo application via Codex Remote, a backend rewrite in Rust in Codex Cloud, and a demonstration with the Hugging Face Microduck robot. OpenAI reports that 1.2 billion people use ChatGPT each week. The keynote concluded at 10:51.