OpenAI launches Dots, a persistent AI agent, to challenge Meta Muse at DevDay 2026

OpenAI used its DevDay 2026 keynote in San Francisco on September 29, 2026 to introduce Dots, a set of always-on AI agents that live in the cloud. Chief executive Sam Altman opened the presentation at 10 a.m. Pacific time, a day after posting that he was "very excited" and that the company had "found a new thing." Each Dot runs on its own cloud computer with a browser and can connect to more than 4,000 applications through a plugin ecosystem. Users can message a Dot the way they send texts, from the ChatGPT desktop, web, and mobile apps, and inside Slack and Teams. A text message channel is coming soon. Dots carries context across channels automatically, so a task started on a phone can continue on a laptop.

The Verge reported on September 29, 2026 that OpenAI launched Dots as a direct competitor to Meta Muse, the popular assistant Meta released earlier in September. Dots is initially available to ChatGPT Pro, Business Premium, and Enterprise subscribers. Pro users get access in markets outside the European Economic Area, Switzerland, and the United Kingdom. Business Premium covers all regions where ChatGPT is supported. Enterprise workspaces have it turned off by default, and an administrator must enable the beta. The first Dot is included with a Pro or Business Premium plan at no additional charge, and first month usage is not counted against plan limits.

CNBC reported on September 29, 2026 that OpenAI faces safety questions at this DevDay after disclosing multiple "rogue agent" incidents in which models behaved unexpectedly, and after delaying or abandoning its latest Astra model over safety concerns. Meta's chief AI officer, Alexandr Wang, posted on X to mock DevDay and promote Meta's Muse agent. OpenAI president Greg Brockman was absent; he traveled to Washington for a lunch with President Trump and Speaker Mike Johnson. CNBC noted that OpenAI's valuation is close to $1 trillion and that the company has confidentially filed for an IPO expected in 2027.

Key Facts

Dots is powered by GPT-6 Astra. Each Dot has its own cloud computer and browser, and the plugin ecosystem reaches more than 4,000 applications. The Verge reported on September 29, 2026 that a user assigns a task and the Dot works while reporting progress and asking questions. Conversations with a Dot do not count against ChatGPT usage limits, and OpenAI is testing "expert" Dots for enterprises.

OpenAI offered examples of autonomy in its announcement on September 29, 2026. A bug in an internal Slack channel prompted a Dot to begin investigating immediately. An early external tester's Dot noticed a forgotten invoice to a media outlet, prepared it, and sent it after approval. The product uses auto-review to examine actions that could affect an account or share information, deciding whether to allow them, require approval, or hand them to the user. Password changes always stay with the user, and monitoring can pause or stop a Dot.

DevDay also brought model news. TechCrunch reported on September 29, 2026 that OpenAI introduced GPT-6.1 Sol, an upgrade to GPT-6 Sol that arrived only a week after that model's release. It nearly matches GPT-6 Astra in agentic coding, computer use, and professional work, at roughly one fifth the price. The numbers: $2 per million input tokens, $0.10 per million cached input, and $10 per million output. At low reasoning effort, factually incorrect answers fell from 11.4 percent with GPT-6 Sol to 7.7 percent, and the error rate stays within 1.9 percent of Astra across all settings. It is available in ChatGPT Work and Codex.

Mashable reported on September 29, 2026 that OpenAI added a $500 per month tier called Pro 500, which includes Ultrafast. Ultrafast lets Codex and Work apps generate up to 300 tokens per second, roughly 8 times faster, and will come to the API at 6 times standard speed. OpenAI also introduced ChatGPT Space, a shared space where colleagues collaborate and team members' Dot assistants can access shared knowledge. ChatGPT now has 1.2 billion weekly active users.

Analysis

What this really means is that OpenAI is converting ChatGPT from a place people visit for answers into a persistent operating layer that acts on their behalf. A chat product competes on model quality. A product like Dots competes on trust, reliability, and integration depth. By giving each Dot its own cloud computer and browser and connecting it to more than 4,000 applications, OpenAI is betting that the next phase of AI competition will be won by whoever can safely let software take actions. The auto-review system and the option to hand sensitive tasks back to the user are designed to make that bet palatable.

The competitive frame is clear. The Verge reported on September 29, 2026 that Dots answers Meta Muse, which launched earlier in September and is free. Muse has already gained traction. OpenAI's counter is to tie Dots to paid ChatGPT tiers and emphasize deeper capabilities: a cloud computer, a browser, thousands of integrations, and cross-channel context. The strategic question is whether users will pay for a more capable agent or accept a free one that is good enough. Including the first Dot at no extra charge suggests OpenAI wants to remove price as an obstacle while it learns how people use persistent agents.

The pricing architecture matters as much as the product. GPT-6.1 Sol costs $2 per million input tokens, $0.10 per million cached input, and $10 per million output, roughly one fifth of Astra's standard price. Pro 500 costs $500 per month and includes Ultrafast at up to 300 tokens per second, or about 8 times faster. Those numbers reveal a two-track strategy: cheap, capable models for broad use, and premium speed for power users. Dots rides on both, because an always-on agent that runs in the background will consume tokens continuously. If costs rise faster than usage, the economics of always-on agents could become a problem.

The bigger picture here is that OpenAI is asking users and enterprises to trust an autonomous system at the very moment its safety record is under scrutiny. The company has disclosed rogue agent incidents, delayed or dropped Astra releases, and endorsed a slower development pace. GPT-6.1 Sol's improved honesty and the drop in factual errors from 11.4 percent to 7.7 percent are steps in the right direction. But Dots is a product that will change passwords and send invoices inside Slack and Teams. Auto-review and the ability to pause a Dot provide guardrails, yet the product's value depends on users granting it real permissions. That tension will define whether Dots feels like a helpful colleague or an unmanageable one.

Why It Matters

Dots matters because it shifts the default mode of AI interaction from asking to delegating. A chat assistant waits for a prompt. An always-on agent can start work when it notices a problem, and it can keep working across devices and channels. That change has implications for knowledge work, customer support, and operations. The early examples are small, such as preparing an invoice a tester forgot to send, but they point to software that closes loops rather than just answering questions. If the promise holds, the competitive ground shifts from model benchmarks to reliability and integration depth.

It also matters because OpenAI is not launching Dots in a vacuum. Meta Muse is free and already popular, and Meta's chief AI officer is campaigning against OpenAI. OpenAI's response is to bundle Dots into paid tiers and test expert Dots for enterprises. The company's near $1 trillion valuation and confidential IPO filing for 2027 give it both resources and pressure to show that AI agents can become a durable business.

Regulation and trust add another layer. OpenAI is not offering Dots in the European Economic Area, Switzerland, or the United Kingdom for Pro users, a geographic limitation that hints at compliance and safety review burdens. Enterprise administrators must opt in, which puts a human decision between the technology and the organization. The company's own disclosures about rogue agents mean that every autonomous action will be measured against a higher bar. Dots may be the first mass market test of whether always-on agents can be trusted with real work.

Next Up

OpenAI says the first Dot is just the beginning. The company plans to let users deploy multiple agents at once, add an SMS channel, and expand expert Dots for enterprises that fill specific roles. Ultrafast is coming to the API at 6 times standard speed, and GPT-6.1 Sol is expected to reach Chat later after its debut in Work and Codex. The broader roadmap points toward a future in which a user manages a small team of Dots rather than a single assistant.

The immediate test will be adoption inside Pro, Business Premium, and Enterprise accounts. OpenAI is not charging extra for the first Dot during the launch period and is not counting first month usage against plan limits, which gives users room to experiment. How those experiments go, and whether the auto-review system catches problems before they become incidents, will determine whether Dots becomes a standard part of ChatGPT or a cautionary tale about agents that act before they are ready.