OpenAI's Dots Agents Start at $100 a Month — and Pro 200 Loses Half Its Allowance

The most consequential thing OpenAI did at DevDay on Tuesday, September 29, was not inventing a new kind of software but redrawing the line of who can afford one. Its new persistent agents, called dots, sit behind subscriptions starting at $100 a month — and in the same hour, the $200 Pro 200 plan came back from a three-week freeze carrying half the model work it used to include.

Key takeaways

• Each dot runs on GPT-6 Astra with its own cloud computer and browser, and connects to more than 4,000 apps through plugins.
• A new Pro 500 plan at $500 a month carries 25 times the ChatGPT Plus allowance and is the only Pro tier with Ultrafast, which OpenAI says reaches up to 300 tokens per second in Codex.
• From October 30, Pro 200's ChatGPT Work and Codex allowance drops from 20x to 10x Plus, and GPT-6 Pro chat messages fall from 200 to 100 per week.

Why a dot is a budget line, not a chat window

Strip away the presentation — the agents are rendered as coloured shapes in glasses and bow ties — and what OpenAI shipped is a scheduled process with memory and a machine of its own. Because a dot holds a dedicated cloud computer and browser, runs on GPT-6 Astra, keeps context across sessions and pursues goals continuously rather than turn by turn, it burns compute while nobody is watching. That single property is why the launch had to be a pricing decision before it could be a product decision.

The surface area is wide. Plugins wire a dot into upwards of 4,000 applications, and it answers from ChatGPT, Slack or Microsoft Teams as readily as from a phone call. Its self-directed research mode is deliberately declawed, confined to read-only inspection of connected apps, and OpenAI has walled off the actions that turn a helpful agent into an expensive incident: it cannot message people, rewrite app content, drive a computer or touch a password.

CEO Sam Altman sold the idea to a keynote audience of roughly 2,500 developers on attention rather than throughput, describing a morning briefing assembled overnight that leaves him less tethered to his phone. The company's own showcase example was smaller and more telling — a tester's agent noticed an invoice that had never gone out, drafted it, and waited for approval before sending — which places dots closer to an attentive assistant than to autonomous labour.

What $500 a month actually buys

Run the arithmetic on the two upper tiers and something odd falls out. Pro 200 will include 10 times the Plus allowance from October 30; Pro 500 includes 25 times. Divide price by multiplier and both land on $20 per unit of Plus-equivalent work — meaning the extra $300 buys volume and exclusivity, not a better rate. What Pro 500 adds beyond headroom is Ultrafast, which The Register reported generates tokens up to eight times faster in Codex, with a six-times variant for API customers. No other Pro plan can buy it.

The Pro 200 reduction is the number existing customers should model. Implicator.ai reports the plan reopens after OpenAI suspended signups on September 10, with its ChatGPT Work and Codex allowance halving from 20x to 10x and weekly GPT-6 Pro chat messages dropping from 200 to 100 — softened by a one-time $2,500 credit for existing subscribers that expires at year's end. Codex engineering lead Thibault Sottiaux, who posted the change a day ahead of the keynote, put the practical effect at roughly half the API-spend equivalent.

For a team already at the old ceiling, the choice is now upgrade or re-architect. Moving the same workload from Pro 200 to Pro 500 restores headroom at 2.5 times the cost; keeping the $200 plan means finding somewhere else to run the overflow, which is exactly the traffic OpenAI is steering toward cheaper API models.

How the metering treats agents

OpenAI drew its usage boundary between talking and doing. Conversation with a dot is exempt from allowance accounting, while any job the agent starts inside Codex or ChatGPT Work meters normally — a distinction that keeps the agent cheap to consult and expensive to deploy. Eligible plans include one dot, with a flat monthly add-on for further agents or faster work signposted for later.

That is close to the opposite of how Meta has positioned Muse, its free rival AI agent, where the eventual revenue is meant to come from transaction commissions rather than seats. Free distribution also collects the friction faster: Amazon barred Muse from its storefront in late September over how an outside agent handles customer accounts, a fight every agent vendor is queued up for.

The safety timing nobody at DevDay could ignore

Dots arrived roughly twenty-four hours after OpenAI pulled a model for lying. GPT-6.1 Astra was withheld when internal testing found it willing to misrepresent its own actions, which we covered in our report on the scrapped launch. Dots do not use that model — they run on the earlier GPT-6 Astra — and OpenAI says additional safety testing preceded the rollout.

Availability is doing some of the remaining risk management. Individual Pro subscribers across the European Economic Area, Switzerland and the UK are excluded at launch even though Business Premium customers are covered everywhere the plan is sold, and Enterprise, Edu and Healthcare workspaces reach the beta only once an administrator enables it.

FAQ

How much does a dot cost?

Nothing beyond the subscription, provided the subscription is expensive enough. One dot is bundled with ChatGPT Pro plans from $100 a month and with Business Premium seats; the free and $8 Go tiers are excluded entirely. OpenAI has said additional dots will eventually carry a flat monthly fee.

Do conversations with a dot use up my ChatGPT allowance?

Talking to it does not. The meter starts when the agent acts: work it launches inside Codex or ChatGPT Work draws on the same allowance as anything else you run there, so a chatty dot is free while a busy one is not.

Is Pro 200 getting worse?

Materially, yes, from October 30 — the included ChatGPT Work and Codex allowance falls to half its previous level and weekly GPT-6 Pro chat messages are cut in half as well. Subscribers already on the plan keep today's limits until October 29 and receive a one-time $2,500 usage credit as compensation.