OpenAI is buying enterprise AI coding accounts with price, not features alone, and the first confirmed switch came from Replit. The developer platform’s chief executive said on October 16 that the company was no longer working with Anthropic. According to The Information, OpenAI offered Replit more than 30 percent off a contract worth over $100 million a year after Anthropic declined to extend a prior discount without a larger spending commitment. OpenAI is now the default model provider for Replit Agent, although seven Claude models remain selectable in the catalog. The move shows how a familiar AI rivalry is changing inside corporate software budgets, where lower inference costs, flexible contracts and default placement can matter more than a provider’s latest model launch.
Why OpenAI Won Replit’s Coding Contract
Replit’s change is not a clean break. The company still supports Anthropic’s Claude models and says it remains model agnostic. Yet the commercial and product center of gravity has shifted. OpenAI is the default, customers must manually select Claude if they want it, and Replit’s own testing reportedly showed stronger results from OpenAI’s Sol and Astra models when building applications from natural-language prompts.
Price was a decisive factor. Replit had been one of Anthropic’s largest customers, but the company received a discount that Anthropic would not continue without a bigger spending commitment. OpenAI used that leverage to offer a reduction of more than 30 percent on an annual contract exceeding $100 million. Even after the discount, OpenAI appears willing to trade some unit economics for a strategic foothold in a market where coding agents process large volumes of requests and customers need predictable bills.
The arrangement also gives Replit strategic flexibility. A multi-model catalog reduces the risk of dependence on one laboratory, while a powerful default lowers friction for the majority of users who never alter model settings. That distinction is crucial because Replit’s enterprise and government business is expanding. Enterprise contracts represented about 2 percent of revenue last year and roughly 15 percent this year, with company executives expecting those customers to account for more than half of revenue in 2027. Default model placement can therefore shape Replit’s costs, its customer experience and its future margins at the same time.
Rippling’s Numbers Show the Migration Is Real
Rippling provides the clearest measure of how quickly usage can move. At the HR software company, OpenAI’s share of coding work rose from 16 percent in May to 52 percent by October. Anthropic’s share fell from 57 percent to 18 percent over the same period. These figures do not describe a small pilot. They show a production workload changing hands in a matter of months, with OpenAI becoming the majority provider for code generation and related engineering tasks.
The shift was not driven solely by a discount. Rippling’s leadership considered economics alongside model capability, but savings on high-volume coding usage can make a lesser operational disruption than changing an engineering team’s preferred tool. A provider that charges less by the token or offers more favorable seat terms can become attractive without requiring developers to learn a new workflow. Once usage moves, evaluation data, internal integrations and purchasing commitments may reinforce the new arrangement.
OpenAI is taking aim at a segment where Anthropic had built an early lead. Claude became closely associated with coding assistants and autonomous software-development agents, and Replit’s reliance on Anthropic helped the platform scale its agent capabilities. OpenAI is now leveraging its broader ChatGPT footprint, developer tooling and enterprise distribution to compete for the same budgets. The contest is shifting from benchmark wins to unit costs and who owns the default setting inside tools used every day.
Discounts Are Reshaping Enterprise AI Economics
DataCamp said OpenAI’s pricing reduced its model costs by 80 to 90 percent. The company also received a reported 40 percent discount to use Codex. A US bank with more than $7 billion in assets doubled its ChatGPT seats to 500 workers at under $7 per seat, compared with a usual price of $20 or more. Those examples suggest that OpenAI’s approach reaches beyond developer platforms and into education, financial services and other businesses trying to expand AI access without increasing software budgets.
There is an important difference between headline discounts and sustainable price reductions. Some savings reflect negotiated enterprise terms, temporary credits or strategic pricing intended to win share. OpenAI can also lower costs through more efficient models and infrastructure, but aggressive offers may still place pressure on gross margins. That creates a difficult but potentially productive dynamic: providers must use better systems to fund lower prices while customers become more willing to test competing models, switch defaults and demand contractual flexibility.
Anthropic’s growth is already encountering pressure in this environment. The company has been valued at nearly 44 percent of coding usage among Ramp’s business customers, compared with OpenAI’s 40 percent, but OpenAI is growing faster in the third quarter. Lower-cost Chinese models from providers including DeepSeek and Kimi add another source of competition, forcing leading US laboratories to reconsider what enterprise customers will pay for premium models.
OpenAI’s Strategy Has Limits
A default is influential, but it is not the same as an exclusive relationship. Replit’s documentation still lists seven Claude models, and a late-September company post said Claude Fable 5 and 5.1 remained in Replit Agent’s production environment. A person familiar with the arrangement also told TechCrunch that the Anthropic partnership remained in place. Replit later clarified that OpenAI was its preferred partner rather than its only provider.
Customers may resist automatic switching when Claude offers stronger performance for a particular repository, architecture or programming task. Developers can also build prompts, evaluation processes and habits around one model family. Switching costs are therefore lowest for generalized coding requests and higher for specialized workflows with extensive internal knowledge. OpenAI’s strategy works best when customers treat models as interchangeable utilities, which is increasingly common in agent-driven software development.
Contract size is another constraint. A discount above 30 percent can determine whether a platform commits to a preferred supplier, but it does not guarantee permanent loyalty. Anthropic could respond with revised prices, stronger enterprise terms or improved product performance. Replit may also use its growing purchasing power to negotiate a multi-year agreement that extracts better terms from several suppliers. The next few renewals will reveal whether OpenAI’s pricing offensive becomes a durable market structure or a temporary share-buying campaign.
What the Coding Market Will Look Like Next
The immediate winners are enterprise buyers. Replit gets lower model costs and retains a choice of providers. Rippling has moved most of its coding workload toward OpenAI, DataCamp reports substantial savings, and a large US bank expanded access to ChatGPT at a sharply reduced seat price. These examples give every organization negotiating an AI contract a new benchmark: if a provider cannot explain why its price is defensible, customers now have credible alternatives.
The broader implication is that enterprise AI competition will increasingly be decided by boring commercial details. Pricing tiers, committed-use discounts, model availability, latency, reliability and default placement can outweigh a dramatic release. OpenAI has already shown that it will sacrifice near-term margin to secure distribution. Anthropic remains capable and relevant, particularly where developers prefer Claude, but its former lead now faces a supplier willing to win customers on price.
For now, OpenAI has changed the terms of the AI coding market. Replit’s default switch, Rippling’s rapid migration and dramatic customer savings are more than isolated procurement wins. They reveal a market in which model quality still matters, but affordability determines who gets used at scale. As more contracts reach renewal, the decisive question will be whether OpenAI can maintain those discounts while improving margins, and whether customers that adopted its models for price remain when the promotional edge fades. OpenAI’s strongest argument is not that every enterprise will use one model forever, but that it can make its models the cheapest and most convenient default for millions of coding tasks.