OpenAI is not winning the AI coding market by building a smarter model this month. It is winning it by cutting the price. A new report from The Information, titled “Anthropic’s Margin is OpenAI’s Opportunity,” describes a deliberate discount campaign aimed at Anthropic’s biggest coding customers, and the early results suggest it is working.
The headline case is Replit. The coding platform moved most of its AI spending from Anthropic to OpenAI under a contract worth more than $100 million a year, after OpenAI offered a discount of more than 30 percent, according to the report. Anthropic, the report says, declined to extend Replit’s previous discount unless the company committed to spending more. Replit chief executive Amjad Masad said on stage at TechCrunch Disrupt in San Francisco on October 16 that the company was no longer working with Anthropic, calling the decision one based on “cost, partnership, and performance,” and adding that “ChatGPT is better now.”
The split is messier than the headline suggests, and that matters. Masad later clarified that Replit “remains model agnostic,” with OpenAI now the preferred partner and default provider for Replit Agent, while customers can still choose Claude. Replit’s own documentation still lists seven Claude models, and a person familiar with the deal told TechCrunch the Anthropic partnership remains in place. What changed is the default, and in developer tools, the default is where most of the spending goes.
Rippling flipped from 57 percent Claude to 52 percent OpenAI
The second case in the report is even starker. At HR software company Rippling, Anthropic’s models handled 57 percent of engineers’ coding work in May, while OpenAI’s handled 16 percent. Those shares have now flipped: OpenAI handles 52 percent and Anthropic 18 percent. Rippling chief executive Parker Conrad said the company switched because OpenAI’s pricing was more competitive.
DataCamp, an online learning platform, is the third named customer. Its chief executive, Jonathan Cornelissen, said OpenAI’s pricing cut the company’s AI costs by 80 to 90 percent. A separate report puts DataCamp’s discount on OpenAI’s Codex tools at around 40 percent, so treat the exact figure as contested, but the direction is not: DataCamp moved, and price was the reason given. A US bank with more than $7 billion in assets reportedly had its ChatGPT seats doubled to 500 employees at no extra charge, keeping its annual contract at about $40,000, or under $7 per user against a standard price of $20 or more.
| Customer | What changed | The number that matters |
|---|---|---|
| Replit | Most AI spending moved to OpenAI | 30 percent-plus discount on a $100 million-plus annual contract |
| Rippling | Coding workload flipped from Claude to OpenAI | OpenAI share 16 to 52 percent; Anthropic 57 to 18 percent |
| DataCamp | AI costs cut sharply after switching | 80 to 90 percent cost reduction claimed; one report puts the Codex discount at 40 percent |
| US bank ($7 billion-plus assets) | ChatGPT seats doubled at the same fee | 500 seats for about $40,000 a year, under $7 per user |
Discounts work because the models are close enough
The deeper point of the report is that price can now move workloads that benchmarks alone could not. When two frontier models can both write working code, a 30 percent discount, or an 80 percent one, beats a few points on a coding benchmark. OpenAI can afford the cuts partly because its newer Sol and Luna model families are cheaper to run, according to the report, which means some of the discount is real efficiency, not just margin sacrificed for market share.
That is not the whole story, though. OpenRouter usage data cited in coverage of the report shows Anthropic regained the top spot after releasing Claude Opus 5.5 last month, with usage running around twice OpenAI’s over the past week. Discounts are winning enterprise contracts, while developers choosing freely still lean toward Claude. Both facts can be true at once, and the tension between them is the real contest: OpenAI is buying the enterprise default, Anthropic is holding the developer preference.
Anthropic has its own pressure. An investor memo cited in the reporting says the company’s growth slowed between the second and third quarters, and both labs face cheaper competition from China, where DeepSeek and Kimi now deliver comparable performance for a fraction of the cost. That squeeze is why this price war matters beyond the two companies: when the leading labs discount this aggressively, every smaller buyer eventually benefits, because list prices become starting points for negotiation rather than fixed rates.
What this means for anyone buying AI tools
If you pay for AI coding tools, three practical lessons follow. First, never accept a list price on an annual contract without asking what a competitor would charge; Replit’s deal shows the discount only appeared when Anthropic refused to match one. Second, watch the default, not the catalog: a tool that still “supports” your preferred model but defaults to another will quietly move your spending. Third, lock in the rate, not just the discount: a 30 percent cut that expires in a year is worth less than a smaller permanent one, and enterprise AI pricing is moving fast enough that today’s deal can look expensive by spring.
Anthropic has not publicly responded to the report, and OpenAI has not confirmed the individual deals. The numbers come from The Information’s reporting and the executives quoted in it, so treat the contract details as reported, not confirmed. But the pattern across three named customers, plus a bank, is now too consistent to dismiss: in the AI coding war, the price tag is doing more damage than any benchmark. For the full price picture, see our October 2026 token price list, our coverage of OpenAI’s models arriving on Google Cloud, and Anthropic’s Claude Duet agent app.
