OpenAI is buying share from Anthropic with discounts large enough to change coding stacks

The model market is becoming liquid enough for price to reverse an enterprise deployment within months. In May, Anthropic handled 57% of the AI coding work across hundreds of Rippling engineers, against 16% for OpenAI. Rippling chief executive Parker Conrad now says OpenAI handles 52% and Anthropic 18%, after the HR-software company sought lower costs and OpenAI proved cheaper.

Replit offers a second, more concentrated example. OpenAI reportedly discounted model prices by more than 30% to win a contract worth over $100mn annually after Anthropic declined to extend its previous price break without a larger spending commitment. Replit chief executive Amjad Masad initially said the company no longer worked with Anthropic, then clarified that OpenAI had become the preferred partner rather than the exclusive one. Claude models remain available, but OpenAI is the default for Replit Agent.

These are customer statements and contract reports, not a controlled model comparison. Workload share mixes price, performance, latency, reliability and internal routing decisions. Rippling's figures also describe one company, and Replit has an incentive to publicize favourable terms. Still, the direction is strategically important: providers can no longer assume that coding customers become captive once developers learn a particular model's behaviour.

Anthropic's historically stronger gross margins gave it room to preserve premium pricing. OpenAI is using its own capital position to challenge that choice account by account. For buyers, the practical lesson is to separate model quality from negotiated economics. Maintain task-level evaluations, route workloads rather than standardising prematurely, and preserve enough portability to reopen pricing when a rival improves. For the laboratories, every discount poses the same question: whether gaining high-volume inference work improves utilisation and distribution enough to compensate for lower margins. The coding market may be revealing that answer before the rest of enterprise AI does.