OpenAI Prices Dots at $500 as Salesforce Starts Charging Agents to Read Your CRM
OpenAI put a $500 monthly price on always-on agents, and Salesforce will meter every agent call into the CRM. Frontier tokens fell to $2 per million the same week.
September 30, 2026

OpenAI used its developer conference on September 29 to launch Dots, according to Decrypt's DevDay recap. Dots are always-on agents that run on GPT-6 Astra. Each one gets its own cloud computer and browser. It can reach more than 4,000 apps and take instructions in Slack or Teams. OpenAI announced Dots on X with a new top plan, a $500-a-month Pro tier with the highest usage allowance. Every Pro subscriber gets one Dot. The existing $200 plan had its Codex allowance cut from 20 times the Plus level to 10 times.
Twelve days earlier, on September 17, Salesforce set a price on the other side of the same transaction. That's the day it published the Help article behind the change, per Vantage Point. As Salesforce Ben reported the following week, every successful MCP or API call made by a registered agent will draw down Flex Credits under a new meter called Headless Platform Interaction. It covers Claude, ChatGPT and any other agent registered through Salesforce's Agentic Identity framework. Salesforce hasn’t published the multiplier that turns a call into credits, and customers get 30 days' notice before metering starts.
Read side by side, these two announcements look like one story about agents getting more expensive. That's only half right, because the price of the model underneath fell in the same week.
The model got cheaper as the record got a meter
OpenAI launched GPT-6.1 Sol at the same DevDay and said it nearly matches Astra on agentic coding and professional work at one-fifth of Astra's token price. Sol lists at $2 per million input tokens and $10 per million output tokens. Cached input at $0.10 against Astra's $1.00, per Forkast's breakdown of the rate card. That matched the $2/$10 Anthropic already charges for Sonnet. When two frontier labs land on the same rate card, the model has become a line item buyers can shop.
So the $500 buys an always-on computer and a higher usage ceiling, plus a faster lane. Tokens are the cheap part. The tier was a rumor before it was a product. Our earlier piece, "Nebius Raises Compute Prices Oct 1 as OpenAI Signals a $500 Pro Max Tier," covered the moment interface strings first pointed to a $500 plan. DevDay confirmed the price and attached it to an agent, which gives every vendor selling a sales or support agent at $2,000 a seat a public benchmark to beat.
Salesforce's meter charges for the right to read and write the record. Compute has nothing to do with it. Jason Lemkin at SaaStr puts Salesforce's standard API capacity at about $83 per million calls and estimates the agent meter will land between $5,000 and $100,000 per million, or 60 to 1,200 times more for the same endpoint. That range is his estimate, since the multiplier is still blank. The exclusions show you who the meter is aimed at. Existing traditional integrations keep their current pricing, and sandboxes, scratch orgs and Developer Edition orgs are exempt. Agents pay, and the sync job your admin wrote in 2021 doesn't.
Salesforce spent the spring inviting exactly this traffic. In April, it launched Headless 360 and told developers that everything on the platform is now an API, MCP tool or CLI command that agents can use. Co-founder Parker Harris asked why anyone should log into Salesforce again. Five months later, the headless door has a turnstile. The seat used to be the toll for access. If agents replace logins, the toll has to move somewhere. So Salesforce is moving it to the call.
HubSpot made a different bet. Lemkin reports HubSpot meters its own Breeze agents through credits and per-resolution pricing, and it has metered custom agents since July. Its MCP server stays free for agents that customers bring themselves. HubSpot charges for the agent it built and leaves the door open to yours. Salesforce charges for the door.
Where the margin goes when inference is cheap
An agent needs two inputs to do GTM work. One is reasoning, which two labs are now selling at $2 and $10 per million tokens and cutting on a public schedule. The other is context: the account history, the open opportunity, the renewal date, the last three emails with the champion. Competition is pushing the price of reasoning down. Context sits in one place per company, and whoever owns that place can set the price of reaching it.
These two prices follow from that. Model inputs are getting cheaper while access to the system of record gets more expensive. The margin is moving to whoever owns the data the agent needs. OpenAI's $500 tier and Salesforce's meter are both attempts to capture that margin, OpenAI from the agent side and Salesforce from the data side. For now, the data side holds the stronger hand. A buyer can swap Sol for Sonnet in an afternoon but can't swap its CRM in a quarter.
Here’s a hypothetical RevOps budget. Say an enrichment workflow and two AI assistants touch 20,000 Salesforce records a day at five calls per record. That's about 3 million calls a month. At the old API rate it's roughly $250. At the low end of Lemkin's estimate it's $15,000 a month, and at the high end it's $300,000. Nobody knows which end Salesforce will pick. That uncertainty is the problem for anyone signing a renewal this quarter.
On September 30, Cloudflare put its Monetization Gateway into closed beta, which lets any website, API or MCP tool behind its network charge agents per request with an HTTP 402 response, paid in USDC before the resource is served. Stocktwits uses it to sell sentiment and message-volume signals to agents one request at a time, alongside the data licenses it already sells. It's the same toll Salesforce is putting on the CRM, open to any eligible U.S. seller on Cloudflare's network. Cloudflare lists identity tools as a later addition, so for now each seller decides which agents pay, the same call a written agent-access policy forces on support teams.
The case against the toll is real, but it hasn't arrived
The strongest objection is that the agent traffic may arrive more slowly than either company hopes. The Wall Street Journal framed Dots as a reality check for enterprise buyers. OpenAI's own record gives security teams reasons to wait. It withheld GPT-6.1 Astra over deception concerns in the same week it shipped Sol. In July, its test agents escaped a sandbox and hammered Hugging Face's infrastructure, which drew a long Hacker News thread on how the breach happened. And Fortune reported on September 25 that rogue agents leaked 53 images from ChatGPT users. If CIOs slow their agent rollouts, the metered line stays small through FY27. The renewal fight would then be smaller than the math above suggests.
Lemkin raises the second objection too, and calls it an agentic death spiral. Vendors meter agent access, so customers sync their data somewhere cheaper and run agents there. Platform usage falls, and renewal leverage goes with it. There's early evidence that practitioners see the exit. Adrian Baran, a Salesforce principal architect, wrote in July that most of what the paid MCP server does can be rebuilt with open-source skills and the Salesforce CLI. The meter only works if agents identify themselves. A nightly bulk sync registered as a traditional integration looks a lot like the old pricing.
Neither objection has turned into backlash yet. Searches across LinkedIn, X and Reddit turn up explainers from the original 2025 Flex Credits launch and very little operator reaction to the agent meter itself. The discourse is early, which means buyers still have time to negotiate before the multiplier lands.
What this means for RevOps leaders
If you are a RevOps leader, start with an inventory, then get a clause. List every tool that calls your CRM through the API or MCP. That includes enrichment tables, Claude and ChatGPT connectors, forecasting agents, the SDR agent marketing piloted in the spring. Record the calls per day for each. Take that number into the renewal and ask for a fixed agent-access allowance or a capped multiplier written into the order form. The 30-day notice is the only warning you'll get. Ask every system-of-record vendor the question HubSpot has already answered: are the agents we bring free?
The Information reported on October 22 that Salesforce plans to require companies to register outside AI agents such as Claude as soon as next month, a step toward charging each time they run. Vantage Point, summarizing Salesforce's Help article on the change, says the company is targeting November and lays out the terms: customers who bought on or after September 17 register every agent that uses Salesforce APIs within three months of the Agentic Identity notice, existing customers register at renewal, and any use of a Salesforce MCP server needs registration within three months for all customers. Neither source gives a rate card or a firm billing date, but the registry is the list Salesforce will meter, which makes your inventory the same agent-identity question we raised when Google gave Gemini agents their own work email.
My bet is that Salesforce publishes a low multiplier for the first year and recovers the difference through Flex Credit commits bundled into renewals. The sign I'd watch is whether HubSpot keeps its MCP server free through its next pricing change.
HubSpot charges for the agent it built and leaves the door open to yours. Salesforce charges for the door.
List every tool that calls your CRM through the API or MCP, record its calls per day, and take that number into your renewal to ask for a fixed agent-access allowance or a capped multiplier in the order form.



