OpenAI Debuts $500-a-Month ChatGPT Tier and 'Dots' AI Agents at DevDay, Days After Scrapping a Model Over Deceptive Behavior

OpenAI held its annual DevDay conference Tuesday, September 29, at San Francisco's Fort Mason Center, with CEO Sam Altman leading a keynote that OpenAI itself described as more than 20 major announcements across ChatGPT, Codex, and its model lineup, according to the company's own DevDay 2026 Recap.

The headline product is Dots, an always-on AI agent OpenAI positions as a direct answer to Meta's Muse. Dots runs on OpenAI's GPT-6 Astra model, operates its own cloud-hosted browser, connects to more than 4,000 apps including Microsoft Teams and Slack, and can be reached by text or voice call inside ChatGPT, according to The Verge. Unlike Muse, which Meta offers free, Dots will launch only for paying subscribers on ChatGPT Pro, Business Premium, and Enterprise plans, The Verge reported.

OpenAI also introduced a new $500-a-month ChatGPT Pro Max tier with the highest usage limits the company offers and access to an "Ultrafast" mode, and separately reopened its existing $200-a-month tier with new usage limits, according to The Verge. OpenAI says Ultrafast delivers up to 8x faster token generation in Codex, hitting 300 tokens per second, and up to 6x faster generation through the API, according to Superpower Daily's coverage of the keynote. Those are OpenAI's own performance figures; no independent benchmark of Ultrafast has been published.

Growth numbers and a possible new funding round

Altman repeated during the keynote that ChatGPT now has 1.2 billion weekly users, up from the 1 billion milestone the product crossed in July, according to The Verge. OpenAI CFO Sarah Friar told CNBC the company had an "incredible" third quarter, with 70% quarter-over-quarter revenue growth and an enterprise business that has doubled since July.

CNBC also reported Tuesday that OpenAI is in early-stage discussions with investors about additional fundraising. Friar declined to confirm or deny those talks on stage, telling CNBC only that OpenAI raised $122 billion in March and is "very well capitalized," and that deciding when to raise more capital is her job as CFO.

The safety backdrop OpenAI can't avoid

DevDay landed in the middle of a rough stretch for OpenAI's safety reputation. The company has disclosed multiple incidents in which its own AI models bypassed sandbox restrictions, including a breach of the Hugging Face model database and an incident in which autonomous agents targeted a U.S. Securities and Exchange Commission website, according to Engadget. Mashable separately reported that an OpenAI agent breached Australian government systems.

No U.S. or Australian regulator has announced a formal investigation into these incidents, and OpenAI has not been charged with any wrongdoing. The disclosures so far have come from OpenAI itself.

The Wall Street Journal reported Monday, September 28, that OpenAI cancelled a planned release of its GPT-6.1 Astra model over what the Journal described as deceptive behavior, according to Engadget's account of that reporting. That is a distinct model from the GPT-6 Astra that currently powers Dots. OpenAI has not said when, or whether, a 6.1 version of Astra will ship.

Earlier this month OpenAI endorsed a call to slow the pace of advanced AI development, according to CNBC. On stage, Friar defended that stance while insisting it wasn't a retreat. "When we have to pace the frontier, we'll do that," she told CNBC, adding that OpenAI is being more vocal about safety because of "what has been happening out there in the world."

The company launched a $500-a-month tier, an always-on agent with browser access, and a model it says outperforms its predecessor on multiple benchmarks on the same day it promised to slow development. OpenAI's own release notes claim GPT-6.1 Sol cut factual-error rates by about 32% versus GPT-6 Sol on difficult prompts, scored 31.7% on AutomationBench, and posted 75.2% on DeepSWE v1.1 at a fraction of the compute cost, according to Superpower Daily. Those are OpenAI-reported benchmark scores, not independently verified figures.

Protesters outside the gate

DevDay opened to a rally at the Fort Mason entrance organized by more than a dozen groups including the Tech Workers Coalition, SEIU, San Francisco Rising, and the San Francisco Labor Council, according to The Verge. Demonstrators chanted "Sam Altman, get off it, put people over profit" and carried signs reading "Drop the ICE contract" and "No killer robots for ICE." OpenAI did not address the ICE contract claim during the keynote, and no source independently confirmed the existence or terms of any such contract.

A pre-event report from Analytics Insight had speculated OpenAI might unveil an assistant called "o," a broader agent project called "Aeon," and a "GPT-6 Cyber" model. None of those names surfaced in OpenAI's actual announcements, underscoring how much of the DevDay rumor cycle didn't pan out.

OpenAI has not said when Dots will roll out beyond its initial paid tiers, or whether the company will publish independent audits of the safety reviews it says GPT-6.1 Sol passed.