Replit Made OpenAI Its Default After a Reported 30% Discount. The Model Price War Is Here. Price Your Product at List

October 25, 2026
7 min read

October 25, 2026
7 min read
For two years, the price of a frontier model was something founders read off a pricing page. This month it became something the largest buyers negotiate, and win. Replit, one of Anthropic's biggest customers, has made OpenAI the default model for its coding agent, reportedly after OpenAI offered a discount Anthropic would not match.
This article is based on The Information's reporting as summarised by Techmeme and The Decoder, and on TechCrunch's coverage of Replit's announcement. Replit, OpenAI and Anthropic have not publicly confirmed the financial terms. The reading from section 3 onward is mine.
At TechCrunch Disrupt on 17 October, Replit's CEO Amjad Masad said Replit was no longer working with Anthropic. He attributed the decision to "cost, partnership, and performance" and said: "I think ChatGPT is better now."
He then softened that. Anthropic's models are still available on Replit, but OpenAI is now the default for Replit Agent. "Replit remains model agnostic," Masad told TechCrunch. "Our users can choose Anthropic's Claude or other supported models." TechCrunch noted that Replit had been a top-five Anthropic customer, and that a person familiar with the partnership disputed Masad's description of it.
A week later, The Information reported what sat behind the switch: OpenAI gave Replit more than 30% off a contract worth over $100 million a year, after Anthropic ended Replit's previous discount. According to The Decoder's summary, Anthropic would not extend that discount without a larger spending commitment.
The same reporting describes a wider campaign by OpenAI to win Anthropic's large customers on price:
The Decoder links part of this to real price reductions from more efficient OpenAI models and infrastructure, and to pressure from cheaper Chinese models such as DeepSeek and Kimi. Some of it is plain customer acquisition.
Replit did not remove Claude. It changed the default. For most users, the default is the product, and the spend follows it.
The detail that matters most is not the discount. It is that Replit kept Claude in its catalogue and moved the default. Most users never open a model picker. Whoever sits in the default slot gets most of the traffic, so changing one setting can move most of a $100 million budget.
That applies to your product too, in both directions:
Price your product at list. A discount tied to volume or a commitment can be withdrawn, as Replit found. Set your own prices so the business works at the provider's published rate, and treat any discount as margin, not as the thing that makes the numbers add up.
Keep switching cheap. Route model calls through one layer, keep prompts and tool definitions per model, and keep an evaluation set built from your real tasks. If moving providers takes a week instead of a quarter, every price change works for you.
Re-run your evals every quarter. Masad's case was partly performance, not just price. Rankings move with each model release. Test the current models from at least two providers on your own tasks and let the results, not habit, set the default.
Ask, even when you are small. You will not get 30% off a $100 million deal. But committed-spend discounts, startup credits and better rate limits are negotiable at much smaller volumes, and a credible alternative is what makes the conversation work.
Know what the routing layer costs you
If you use a router or gateway to stay portable, that layer becomes its own dependency. We looked at what that means when Stripe bought OpenRouter, and at how teams split work across several models to cut costs.
The exact terms. The 30% and $100 million figures come from The Information citing a person familiar with the deal. None of the three companies has confirmed them.
That Claude is now worse. Masad said "ChatGPT is better now" for building apps from scratch on Replit. That is one company's judgement on its own workload, not a general benchmark. Run your own.
That the low prices will last. Discounts aimed at winning customers from a rival are a sales tactic. They may continue, or they may end once the customers have moved.
That Anthropic is losing overall. These are reports about a handful of large customers. They say nothing reliable about either company's total business.
Replit made OpenAI the default for Replit Agent after OpenAI reportedly offered more than 30% off a contract worth over $100 million a year, and after Anthropic ended Replit's discount. Claude is still on the platform, but the default is where the spend goes. Rippling and DataCamp are reported to have moved large shares of their spend as well.
For founders, the price war is good news only if you can act on it. The companies benefiting are the ones that could move their default in days. The ones exposed are those whose margins depended on a discount someone else controls.
This quarter, check that your product is profitable at list price, put every model call behind a single routing layer, and run your evaluation set against a second provider so you know what switching would cost before you need to.
Sources: Techmeme, summarising The Information's report that Replit moved most of its AI spend from Anthropic to OpenAI, which gave more than 30% off a deal worth over $100 million a year, after Anthropic discontinued its discount. The Decoder, 25 October 2026, citing The Information — the Rippling, DataCamp and US bank figures and the context on model efficiency and Chinese competitors. TechCrunch, 17 October 2026 — Masad's remarks at Disrupt, his clarification that Replit remains model agnostic with OpenAI as the default, Replit's status as a top-five Anthropic customer, and the dispute from a person familiar with the partnership. RuntimeWire — the note that the companies have not confirmed the financial terms. The reading in sections 3 to 5 is mine.
IdeaToMVP Academy
4-week live cohort for founders. Learn to ship AI agents, scope MVPs, and automate your business — taught by the same team that writes these guides.