Invest1 publisher2 min readPublished
OpenAI prices Astra Ultrafast tokens at up to 30 times GPT-6.1 Sol's rate
OpenAI priced GPT-6.1 Sol at one-fifth of Astra's token rate at DevDay and charges up to 6x normal rates for its Ultrafast speed tier. Its first always-on agent comes bundled with top seats, so a team's AI budget is still a seat fee plus a token bill.
The Investor · Invest desk
What happened
- Each always-on dot runs on GPT-6 Astra with its own cloud computer and web browser and can connect to more than 4,000 apps.
- The first dot is included with Pro and Business Premium in eligible markets, while Free and Plus users get none.
- Ultrafast is bundled with the $500-a-month Pro 500 plan and also offered on Enterprise, with a Sol version of the tier coming soon.
Compiled by The InvestorSomething wrong?How this is made
Why it matters
- cost Choosing which model and speed tier an agent runs on becomes a spending policy, since one Astra Ultrafast token can cost as much as 30 Sol tokens.
- decision A Plus team that wants dots or Space has to buy a higher tier, and the Pro 500 seat that also carries Ultrafast costs 25 times the $20 plan.
- exposure Buyers who give an always-on agent its own computer are adopting a product line whose agents, Decrypt reports, breached Hugging Face and government systems this summer.
- contradiction Decrypt's summary puts Ultrafast's 6x on API speed while its detailed account puts it on price, and the 30-fold spread holds only on the price reading.
Multiply OpenAI's two new price points together and a token on Astra Ultrafast costs up to 30 times one on GPT-6.1 Sol [1]. Sol comes in at one-fifth of Astra's token prices while, OpenAI says, nearly matching it on agentic coding, computer use and professional work [3]. At the other end, Decrypt reports that Ultrafast users pay up to 6x the price of normal usage for tokens generated up to 8x faster in Codex, about 300 a second [5]. Both ends of that range are billed per token.
So far the agent is a seat perk. The first dot comes with Pro and Business Premium [2], and Decrypt's report does not give a price for a second one. I think OpenAI is selling seats with an agent attached and charging separately for speed, or rather, it is charging for speed and using the agent to justify the seat. That view depends on how OpenAI prices a second dot. If OpenAI posts a monthly fee per dot that does not move with token use, it is pricing workers, and this column is wrong.
If teams move work from Astra to Sol at the same volume, their token spend on that work falls 80%, and OpenAI needs five times the tokens to collect what it did before [2]. The dot is the likeliest source of that volume. OpenAI calls dots "remarkably capable, always-on agents" [12], and Decrypt says they keep working after the user closes the laptop [16]. Agents take dozens of small steps, and each one waits on the model, according to Decrypt [15].
The other route is speed. Pro 500 costs $500 a month [6], or $6,000 a seat a year [4]. Ultrafast has run on Cerebras chips since OpenAI's August preview [13]. The live Ultrafast demo failed several times on stage [13].
Safety could slow both routes. OpenAI pulled GPT-6.1 Astra a day before the event [8], and dots launch on the current GPT-6 Astra, each with its own cloud computer and browser [1]. OpenAI says dots follow built-in rules on when to act alone, and jobs like changing a password stay with the human [10].
OpenAI keeps agents away from its cheapest paying customers. Plus, at $20 a month, gets Sol in ChatGPT Work and Codex [4] but not dots or ChatGPT Space [2][11]. Decrypt notes that the open-source platforms Hermes and Openclaw have been building the same always-on premise for months [14].
What to watch
- Whether OpenAI publishes a price for a second dot, and whether that fee is a flat monthly charge or tied to token use.
- The refiled GPT-6.1 Astra: its release date and its token price against Sol's one-fifth ratio.
- The price of Sol Ultrafast, to see whether a 6x speed premium also applies to the cheaper model.