OpenAI undercut Anthropic on a Replit contract of more than $100 million a year, The Information reports
The Saturday story is paywalled past the first customer. In the free portion, Rippling’s Parker Conrad says OpenAI’s share of his engineers’ coding work rose from 16 percent in May to 52 percent, and Anthropic’s fell from 57 percent to 18 percent. Relays of the rest put Replit’s discount above 30 percent. A 90 percent figure in a Sunday headline is a chief executive’s estimate of a deal he was still seeking.
Deep dive4 minIn-depth
Maya Chen4 min read
- OpenAI
- Anthropic
- Replit
- Rippling
- Pricing

Photo: The InformationSource page
OpenAI has been taking coding work from Anthropic by charging less, The Information reported on Saturday in a story by Laura Bratton and Stephanie Palazzolo. The headline on the piece is “Anthropic’s Margin is OpenAI’s Opportunity.” Most of it sits behind a paywall. The paragraph carried on The Information’s site, and in a Benzinga syndication timestamped 10:00 a.m. Eastern, is about one customer. Parker Conrad, chief executive of the HR software company Rippling, said that in May Anthropic’s models did 57 percent of the coding work of hundreds of engineers there, and OpenAI’s models did 16 percent. Those shares have reversed. OpenAI now handles 52 percent and Anthropic 18 percent, he said, as the company sought to cut costs and OpenAI proved cheaper.
The two named shares do not add up to the whole. In May they account for 73 percent of the coding work Conrad described. In the later figures they account for 70 percent. The reports do not say who does the rest. They also do not say what “coding work” counts: prompts, tokens, or tasks. Conrad is quoted on the direction and on the reason, which is price. He is not quoted on a contract value for Rippling.
A Replit discount above 30 percent, reported rather than published
AI Chat Daily, in an account timestamped 10:09 p.m. UTC on Saturday, reports the Replit terms from the same Information story. OpenAI cut its price by more than 30 percent on a contract worth more than $100 million a year, after Anthropic declined to extend a discount Replit already had unless Replit committed to spending more. The Decoder’s Sunday piece, by Matthias Bastian, gives the same Replit terms and attributes them to The Information. Neither OpenAI nor Anthropic has published the contract. The write-ups do not say whether the $100 million is measured before the discount or after it, so a dollar saving cannot be calculated from the sentence. AI Chat Daily said the reports it read included no statement from Anthropic on the Replit contract or the Rippling shares.
The Replit price lands on a switch the company has already described in public. On 16 October, at TechCrunch Disrupt, chief executive Amjad Masad said OpenAI was the preferred partner for Replit Agent and that customers could still choose Claude. ALFW, in a piece dated 25 October, wrote that it checked Replit’s model-selector documentation that day and still found seven Claude models listed, across what Replit calls Power and Max modes. A default moves the money if most customers leave it alone. A default is not the same thing as Claude leaving the product.
The 90 percent is an estimate of a deal still being asked for
Bastian’s headline says OpenAI is offering discounts of up to 90 percent. ALFW’s reading of the relays, including a Sina Finance account it cites, separates three numbers. What The Information reported as granted at DataCamp, ALFW writes, is a 40 percent reduction on models used for internal product development, conditional on DataCamp also using Codex. Jonathan Cornelissen, DataCamp’s chief executive, estimated that costs could fall by 80 to 90 percent if comparable terms were extended to the models inside the products DataCamp sells. He was seeking that extension. ALFW says the reporting it reviewed does not establish that OpenAI offered, or that DataCamp signed, a 90 percent rate. The Decoder attributes an 80 to 90 percent cost cut to Cornelissen without that condition. The two Sunday pieces are not describing the same fact.
The Decoder adds a third customer the other accounts do not put in the same detail: an unnamed US bank with more than $7 billion in assets, whose ChatGPT seats were doubled to 500 employees at no extra charge, at an effective price under $7 a user against a usual price the Decoder puts at $20 or more. ALFW notes that the institution, the agreement and the billing terms have not been publicly disclosed. A seat expansion on unpublished terms is not a rate card.
Bastian’s closer is his own frame for why OpenAI can cut. He writes that the company is partly buying share at the expense of its margins, and that it has also cut prices in earnest through the Sol and Luna lines, because of more efficient models and infrastructure. He writes that Anthropic’s growth slowed from the second quarter to the third, according to an investor memo, and that both companies face price pressure from new DeepSeek and Kimi models. ALFW’s caution on those two points is that they are secondary reports, not audited figures. They do not say what a buyer will be offered on the next renewal. What the Saturday reporting does put on the record is narrower, and more usable: at Replit, a discount above 30 percent arrived after an older one was not extended, and at Rippling the share of coding work moved with the price.


