Replit just handed OpenAI one of the clearest wins yet in the AI coding wars — and it came directly out of Anthropic’s pocket. According to The Information, Replit recently moved most of its AI spending from Anthropic to OpenAI, in a contract reportedly worth more than $100 million a year, after OpenAI offered a discount of more than 30% and Anthropic declined to extend the discount Replit previously had.
Replit CEO Amjad Masad confirmed the shift publicly on October 16 at a San Francisco conference, saying the company is “no longer working with Anthropic” as its primary model provider. He later clarified that Replit “remains model agnostic,” but the practical effect is unambiguous: OpenAI is now the default provider for Replit Agent, and customers who want Claude models have to select them manually.
The timing matters. Anthropic is reportedly targeting a public listing as early as mid-November, before Thanksgiving — and a flagship coding customer walking out the door, even partially, is not the kind of headline a company wants during IPO season. We covered the financials behind that listing in our breakdown of Anthropic’s leaked IPO numbers, including the roughly $450 billion in compute commitments Anthropic is on the hook for regardless of who’s buying its models.
The 30-Second Version
- The deal: Replit shifted most of its AI spending from Anthropic to OpenAI in a contract reportedly worth $100 million-plus per year, after OpenAI discounted its models by more than 30% (The Information)
- Why: Replit CEO Amjad Masad cited cost, partnership, and performance — he said internal testing showed OpenAI’s models performed better at building applications from scratch
- Anthropic’s move: Anthropic reportedly declined to extend Replit’s existing discount unless Replit committed to spending more, effectively pricing itself out
- Not isolated: At Rippling, OpenAI’s share of engineers’ coding work rose from 16% in May to 52% now, while Anthropic’s fell from 57% to 18%. DataCamp separately got a 40% discount from OpenAI for its Codex models
- Still hedging: Replit’s documentation still lists seven Claude models, and a late-September company blog post said Claude Fable 5 and 5.1 remain in production — OpenAI is the default, not the only option
- Context: The shift lands weeks before Anthropic’s reported IPO target, adding pressure at a sensitive moment
Why Replit Switched from Anthropic to OpenAI
Replit’s public story and the leaked contract details point to the same conclusion from two different directions.
On stage, Masad framed it as a straightforward business decision: “It was a decision based on both cost, partnership, and performance.” He said Replit’s own benchmarking found that OpenAI’s newer coding-focused models — reported elsewhere as GPT-5.6 Luna and the Astra-era models — performed better than Claude at building “vibe-coded” applications from scratch, the kind of from-nothing app generation that is core to Replit Agent’s pitch.
Behind the scenes, The Information’s reporting (via coverage from AI Chat Daily and RuntimeWire) adds the money side: OpenAI cut its price by more than 30% on a Replit contract now worth over $100 million a year. Anthropic, for its part, had reportedly extended Replit a discount in the past, but declined to renew it unless Replit agreed to spend more — a negotiating stance that, in hindsight, backfired. Neither Anthropic, OpenAI, nor Replit has publicly confirmed the exact financial terms; the $100 million-plus figure and the 30% discount are both reported, not confirmed by the companies involved.
Until recently, Replit was described as one of Anthropic’s top five customers by revenue. Replit had announced its OpenAI partnership back in August, but the move described this month is being reported as a much larger reallocation of spend — not just an additional vendor, but a change in which model is the default.
It’s Not Just Replit
The more important part of this story isn’t Replit alone — it’s the pattern around it.
At Rippling, the HR software company, Anthropic’s models reportedly handled 57% of the coding work done by hundreds of engineers back in May, against 16% for OpenAI. The Information’s sourcing says those numbers have since reversed: OpenAI now handles 52% of that work, Anthropic 18%. That’s not a gradual drift — it’s a near-total flip in a matter of months.
DataCamp is reported to have secured a discount of around 40% from OpenAI to use its Codex coding models, another data point in what looks like a deliberate, aggressive push by OpenAI to win enterprise coding workloads away from Anthropic specifically, rather than growing the category generally.
Anthropic has built much of its enterprise reputation on Claude being the preferred model for coding — Claude Code and the broader developer ecosystem around it have been central to that story. A visible defection at Replit, paired with Rippling’s numbers and DataCamp’s discount, suggests OpenAI is specifically targeting that reputation with price as the weapon, not just better benchmarks.
None of the three companies have commented publicly on the Rippling or DataCamp figures, and The Information’s reporting relies on sources described as familiar with the matter rather than named company statements. Treat the specific percentages as reported figures that could be revised, not confirmed audited numbers.
It’s also worth noting what this isn’t. Enterprise coding contracts are only part of Replit’s business — the company has said enterprise and government deals made up roughly 2% of its revenue last year and about 15% this year, with more than half of current revenue coming from business customers generally. Replit’s CEO has said he expects enterprise and government customers to pass half of revenue in 2027. So the OpenAI deal is tied to a business line that’s growing fast, not Replit’s entire company, but it’s exactly the kind of fast-growing enterprise line that AI vendors are fighting hardest to own.
Why Price Wars Happen Now
Price discounting this aggressive usually shows up when a market is maturing from “whichever model is smartest” to “whichever model is cheapest at acceptable quality” — and coding assistance is arguably the first AI use case mature enough to get there. Coding tasks are easier to benchmark objectively than open-ended chat, usage volumes are high and predictable, and switching costs are lower than they look, since most coding platforms (Replit included) are built to swap models rather than lock into one.
That combination is exactly what makes coding workloads the first real battleground for margin, not just capability. OpenAI cutting prices by 30–40% to take share from Anthropic only makes sense if the company believes the long-term value of owning developer habits outweighs near-term margin — a bet that mirrors how cloud providers have historically discounted compute to lock in platform loyalty.
What This Means for Anthropic’s IPO
Anthropic is reportedly aiming to begin formal IPO marketing around November 9, with trading potentially starting before Thanksgiving, at a valuation reports put above $2 trillion. Enterprise coding demand — and Anthropic’s reputation as the go-to model for serious software work — is a meaningful part of the story investors are being asked to buy into.
A price war it’s visibly losing on at least one high-profile account, right as that pitch is being made, is an awkward coincidence at best. It doesn’t mean the IPO is in trouble; Anthropic’s reported revenue growth (covered in our IPO numbers piece) has been dramatic, and losing share on price to a competitor willing to discount more aggressively is a normal competitive dynamic, not a sign of a broken product. But it does undercut the narrative that Claude is simply the default choice for coding, which has been one of Anthropic’s clearest competitive advantages.
What It Means for You
- If you run a coding-heavy AI budget, this is a reminder that discounts at this scale are negotiable — Replit, Rippling, and DataCamp all appear to have gotten better terms by being willing to switch or threaten to
- If you’re choosing between Claude and OpenAI’s coding models, price gaps this large are temporary competitive tactics, not necessarily a verdict on quality — benchmark your own workload rather than following headlines
- If you’re watching Anthropic’s IPO, enterprise coding share is one of several metrics worth tracking alongside the raw revenue and loss figures; a single account loss isn’t fatal, but a trend would be
- If you build on either platform, staying multi-model, the way Replit says it intends to despite the shift, is the safer long-term position
FAQ
Did Replit stop using Claude entirely? No. Replit’s documentation still lists seven Claude models, and the company says Claude Fable 5 and 5.1 remain in production. OpenAI is now the default provider for Replit Agent, not the exclusive one.
How much is the Replit-OpenAI deal worth? The Information reports it at more than $100 million a year, with a discount of more than 30% on OpenAI’s standard pricing. Neither company has publicly confirmed these figures.
Why did Anthropic lose the account? Reported reasons include Anthropic declining to renew Replit’s existing discount without a bigger spending commitment, plus Replit’s internal testing favoring OpenAI’s coding models on performance.
Is this just about Replit, or is it bigger? It appears bigger. Rippling’s reported internal coding-workload share flipped from 57% Anthropic/16% OpenAI in May to 18% Anthropic/52% OpenAI now, and DataCamp reportedly got a 40% OpenAI discount of its own.
Does this affect Anthropic’s IPO? Not directly or necessarily materially, but it complicates the narrative that Claude is the default enterprise coding model, right as Anthropic is reportedly preparing to market its IPO as early as November.
The Bottom Line
OpenAI didn’t just win a customer — it won a customer that had been one of Anthropic’s top five accounts, using price as the primary lever, and it’s reportedly doing the same thing at other companies at the same time. Anthropic still has real strengths and real revenue growth heading into its IPO, but “the best model for coding, and everyone knows it” is a harder claim to make when the receipts show otherwise at Replit, Rippling, and DataCamp. Expect more of these numbers to surface as Anthropic’s roadshow gets closer — and expect Anthropic to respond on price, not just on benchmarks.
Sources
- AI Chat Daily: OpenAI used 30%-plus discounts to win Replit’s $100M AI spend from Anthropic
- RuntimeWire: Replit shifted most AI spending to OpenAI under a reported $100M-a-year deal
- Techmeme: Source citing The Information on the Replit-Anthropic-OpenAI shift
- Our coverage: Anthropic’s IPO Numbers Leak
