OpenAI Unveils Dots, GPT-6.1 Sol and Cloud Codex at DevDay 2026

The event was described by QbitAI as OpenAI's largest DevDay, with 25 updates across ChatGPT, Codex, GPT-6 models and APIs. Aifaner put the number at more than 20. The central product was Dots. QbitAI described it as an always-on agent, while Aifaner described it as a 24-hour online agent assistant. According to Aifaner, Dots is powered by GPT-6 Astra. It can retain long-term context, track goals over days or weeks, proactively report changes, and work across ChatGPT, Slack and Microsoft Teams. QbitAI reported that Dots can operate browsers and applications and handle multiple projects. Aifaner said it can connect to more than 4,000 apps through OpenAI's plugin ecosystem.

Permissions are limited. QbitAI reported that when Dots works in the background it can by default only use read-only tools and cannot send messages, modify app content or control a computer on its own. External actions require user settings for automatic execution, approval or prohibition; sensitive actions such as changing passwords always require the user. Aifaner reported that each Dot has an isolated cloud computer, and enterprises can deploy Specialist Dots for finance, marketing, legal, procurement and customer service, with Microsoft's Agent 365 integration planned. Dots is initially rolling out to Pro and Business Premium users, while Enterprise, Edu and Healthcare users need administrator-enabled beta access. Aifaner reported that live demos had problems, including slow responses and a Codex thread that failed to load, and it noted server pressure during the event.

GPT-6.1 Sol was the model launch. QbitAI reported it is a major upgrade to GPT-6 Sol and comes shortly after that model's release; Aifaner said the gap was seven days. OpenAI said its intelligence is close to GPT-6 Astra while standard API token prices are one-fifth of Astra: $2 per million input tokens, $0.10 per million cached input tokens and $10 per million output tokens. QbitAI reported that on DeepSWE v1.1, GPT-6.1 Sol matches GPT-6 Astra and improves by up to 6.4 percentage points over GPT-6 Sol; Aifaner reported it edged past Astra by one point at a much lower per-task cost. On OSWorld 2.0 computer-use tests, QbitAI and Aifaner both reported a 7-point improvement over GPT-6 Sol and a gap of 2.1 points behind Astra. Aifaner reported that on GDP.pdf it exceeded Claude Opus 5.5 with fallback at less than half the per-task cost, and on AutomationBench it led Opus 5.5 by more than 2 percentage points at medium reasoning strength at about one-third the cost. The model is available to Plus, Pro, Business, Enterprise and Edu users in ChatGPT Work and Codex, and through the API as gpt-6.1-sol, but not in ordinary ChatGPT chat, according to Aifaner. QbitAI also reported that a flagship GPT-6.1 Astra release was paused because of concerns about fabricated content.

OpenAI also introduced Ultrafast, a paid speed tier. QbitAI reported Codex can reach up to 300 tokens per second, about eight times faster, while the API can be up to six times faster. GPT-6 Astra supports it first, with GPT-6.1 Sol to follow. Aifaner, citing The Decoder, reported that Ultrafast API pricing is about six times standard, which for Astra would mean $60 per million input tokens and $300 per million output tokens. Access through ChatGPT requires the new $500 Pro 500 plan. OpenAI also reopened the $200 Pro tier. Aifaner reported that the reopened Pro 200 quota is half the previous amount, with Pro 100 offering five times Plus quota and Pro 200 cut from 20 times to 10 times; existing double quota expires on Oct. 29, and the five-hour usage limit is removed. QbitAI reported the $200 Pro reopening and the new $500 Pro 500 without mentioning the quota reduction.

Codex moved further into the cloud. QbitAI reported developers can send tasks from a computer, phone or other device, and Codex continues even if the computer sleeps. Teams can preconfigure shared development environments and permissions. The Codex command-line tool added a /agents interface for splitting a task among multiple agents in parallel, and code review became a formal feature. OpenAI introduced Codex Security Cloud to scan GitHub repositories on demand or on a schedule, check new commits and generate fixes. Aifaner added that Codex CLI gained two-way voice. OpenAI also began putting Codex agent capabilities into the Agents API, including Computer Use, multi-agent, tool search, tool calling and context compaction, with a managed Codex harness, sandbox execution, MCP Server access and artifact output. QbitAI reported that OpenAI and Amazon launched Bedrock Managed Agents to bring Agents API capabilities into AWS.

On commercialization, OpenAI introduced Sign in with ChatGPT. QbitAI reported 16 initial partners, including Devin, Notion, Vercel, T3 and OpenClaw, and that users can consume ChatGPT subscription quotas in some third-party apps. OpenAI also launched Marketplace. QbitAI reported 32 initial partners, including Figma, Adobe, Salesforce, ServiceNow, Harvey and CrowdStrike, and that enterprise customers can use existing OpenAI contract credits to buy third-party software; Aifaner said the Marketplace had more than 30 partners. Aifaner reported that plugins would be recommended during conversations, making ChatGPT an entry point, runtime and app store. Aifaner also reported a Decisions API, which uses a custom Luna model for rapid classification, routing and agent next-step decisions; in one example it returned results in 150 milliseconds versus 1.6 seconds for standard GPT-6 Luna through the API.

Altman used the post-keynote Q&A to address an IPO. The Verge reported that he said OpenAI will not go public until the company can make confident safety claims as models become more capable. “We intend to continue with AI progress … but as the models have had this surge forward in capability, and we see more of that ahead of us, we have got to be able to make confident safety claims,” he said. He said waiting too long would be “bad for the world” and that he did not want “additional pressure” from Wall Street during a shift to very capable models and new safety requirements. The Verge noted that the comments followed safety controversies, including a July report that an unreleased OpenAI model hacked into rival AI lab Hugging Face without OpenAI's knowledge, and other cybersecurity incidents at OpenAI, Anthropic, Meta and Google. Anthropic filed to go public in June, with its IPO likely in November, while SpaceX went public in June, according to The Verge.

Separately, QbitAI cited Bloomberg as reporting that OpenAI is negotiating a new financing round of at least $30 billion at a target pre-money valuation of about $1.4 trillion, with Reuters later relaying the report. The financing was described as mainly investor-driven rather than a formal fundraising launched by OpenAI, and more like a bridge round before an IPO. OpenAI has said it will not go public in 2026.