OpenAI DevDay 2026: GPT-6.1 Sol, Dots Agents, No New Astra

What OpenAI actually shipped

OpenAI's DevDay 2026 centered on agents and cheaper capability, not a new flagship. The headline model was GPT-6.1 Sol, an update to GPT-6 Sol aimed at coding, computer use, and professional work. OpenAI says it approaches GPT-6 Astra on several evaluations while costing one-fifth of Astra's standard input and output token prices, with cached input at $0.10 per million tokens.5 Another recap also calls Sol the event's major model release and frames it as near-Astra capability at a much lower price.3

That leaves an obvious gap. Nothing in the coverage points to a GPT-6.1 Astra. Instead, the top-tier model stays at GPT-6 Astra, which powers OpenAI's new always-on agents.3 On stage, Sam Altman said the new Ultrafast tier is available "for Astra 6 today, and Sol 6.1 soon."1 Why Astra did not get a point release is unclear. OpenAI offered no explanation in the material covered here, and there is no documented evidence of a cancelled Astra update or safety failures behind one. Any claim along those lines goes beyond the public record.

Dots and the agent push

The product Altman pitched as a "whole new way of working with AI"1 appears to be Dots: always-on agents powered by GPT-6 Astra, each with its own cloud computer, browser, persistent working context, and access to connected apps.3 Simon Willison guessed from the audience that it would be a personal agent along the lines of Meta's Muse.1

The Verge, writing before the launch, said OpenAI's agent effort would need to combine the usefulness of open-source OpenClaw, the low cost and easy setup of Muse, and the conversational simplicity of newcomer Instinct to compete.2 Its piece called the product Aeon, which differs from the "Dots" branding in the post-event recap.23 The Verge's bigger point was security. It noted a now-patched vulnerability that let attackers take over accounts, and user complaints that Muse shared a home address without permission and read private messages. It added that such anecdotes are hard to verify.2

That context matters for Dots. An agent that runs continuously, browses on its own, and holds credentials to connected apps faces the same risks The Verge described, with more exposure. OpenAI's choice to run Dots on Astra rather than the cheaper Sol suggests it wants its most capable model handling unsupervised work. Whether that reduces or increases risk is something the coverage cannot answer yet.

Developer tooling: computer use, Codex, Ultrafast

For builders, the Agents API now supports computer use, so applications can operate software through graphical interfaces.53 It also gains multi-agent features from Codex, tool search, tool calling, and context compaction, with OpenAI managing the execution infrastructure.5

Codex got a large share of the updates:

• Cloud environments: Codex can now run remotely, so tasks can be started from other devices.5
• CLI updates: voice input and an /agents interface for delegating and monitoring parallel tasks.5
• Code review: a workflow that checks GitHub pull requests and GitLab merge requests.5
• Codex Security Cloud: scans repositories and new commits, removes duplicate findings, and drafts fixes.5

Codex on Linux and on phones drew applause in the room.1 Willison noted that he tried to build his live-blog photo tool with Codex Cloud on the way to the venue, ran into problems, and switched to Claude Code for web.1

Ultrafast is a premium inference tier that OpenAI says is about 8x faster, reaching up to 300 tokens per second. It costs six times the standard price and is offered across the API, ChatGPT, and Codex.1 It targets latency-sensitive agent workflows.3

The Decisions API's thin paper trail

The Decisions API, an endpoint for choosing between options defined in advance, was listed among the launches.35 An independent check as of October 2 found almost nothing behind it. Docs and API reference pages returned 404 errors, and there were no changelog or pricing entries, no standalone announcement, and no matching model ID.4 The timing also raised questions. TypeSafe launched Jev, a model built for typed decisions, on September 15, two weeks before OpenAI's version appeared. One Hacker News commenter called the launch validation for Jev.4

The reading

DevDay 2026 showed OpenAI spreading capability downward and outward. It cut prices with Sol, charged extra for speed with Ultrafast, and made agents persistent with Dots. The lack of an Astra point release is notable, but the evidence does not explain it, and claims about why it is missing should be treated with caution. The more concrete open questions are whether always-on agents can avoid the security problems already seen in Muse, and whether announcements like the Decisions API become documented products.