The Information reported on October 24 that Replit — one of Anthropic’s own marquee case studies for Claude Code — moved most of its AI spend to OpenAI this year, in a deal worth more than $100 million annually, after OpenAI offered a discount above 30% and Anthropic wouldn’t match it. The story picked up fast on Techmeme, and for good reason: this isn’t a benchmark argument. It’s a company that builds AI coding tools for a living, running the numbers on two frontier labs, and picking the cheaper one.

That’s a different story than the one Replit CEO Amjad Masad told a week earlier. At TechCrunch Disrupt on October 16, he said flatly that Replit was done working with Anthropic, and that “I think ChatGPT is better now.” Within a day he’d softened that to “model agnostic,” and a source close to the Anthropic relationship told TechCrunch the partnership was still active. The Information’s reporting threads that needle: Replit didn’t blow up the Anthropic relationship. It reallocated the budget, and the reason wasn’t performance. It was price.

Quick Summary: What Happened

Detail	Info
Reported	October 24, 2026, by The Information
Deal size	$100M+/year in AI spend, now mostly routed to OpenAI
OpenAI’s offer	30%+ discount on model pricing
Anthropic’s move	Declined to extend Replit’s existing discount without a bigger spending commitment
Replit Agent default model	OpenAI; Claude requires manual selection (7 Claude models still listed in Replit’s docs)
CEO’s public framing	Amjad Masad, at TechCrunch Disrupt: “ChatGPT is better now” — then walked back to “model agnostic”
Comparable pattern	At Rippling, OpenAI’s share of coding work grew from 16% to 52% (May–Oct 2026) while Anthropic’s fell from 57% to 18%

Bottom line: Anthropic lost its biggest discount fight of the year not on capability, but on margin. It chose not to defend a price point, and a flagship customer walked to the competitor that would.

What Actually Happened

Start with the plain mechanics, because they matter more than the CEO soundbite that got all the headlines. Replit had a discount from Anthropic. Anthropic declined to extend it without Replit committing to spend more. OpenAI, meanwhile, offered Replit a discount of more than 30% to win the business — on a contract The Information pegs at upwards of $100 million a year. Replit took the better number.

That’s it. That’s the whole mechanism. No benchmark war, no secret capability gap discovered in production — a vendor refused to defend its price, and a competitor undercut it hard enough to move a nine-figure account.

Masad’s TechCrunch Disrupt comments dressed this up as a verdict on model quality: OpenAI’s Sol and Astra, he said, “consistently perform better on building vibe-coded applications,” and the decision came down to “cost, partnership, and performance.” Maybe all three are true for Replit’s specific workload. But cost is doing most of the work in that sentence, and the walkback within 24 hours — Replit “remains model agnostic,” per a person familiar with the account — tells you Masad said more than the relationship could bear. Replit Agent now ships with OpenAI as the default model. Claude isn’t gone. It’s seven models deep in a settings menu, there if you go looking for it, no longer the thing you get by default.

Why Would Anthropic Let a $100M Account Walk?

1. Anthropic had the discount and chose not to renew it as-is. Replit wanted the existing terms extended. Anthropic countered with a bigger spending commitment as the price of keeping the discount — a negotiating position, not a concession.
2. OpenAI was willing to buy the account at a loss-leading margin. A 30%+ discount on a $100M+/year contract is a real subsidy, and OpenAI extended it specifically to take share from Anthropic’s customer base, not just to win new business.
3. Anthropic is reportedly managing margin more carefully than OpenAI right now. The Information’s framing — margin as Anthropic’s constraint, opportunity as OpenAI’s lever — suggests Anthropic is unwilling or unable to match discount-for-discount on its highest-volume accounts.
4. Replit’s workload is price-elastic. Coding-agent token volume scales directly with usage, so a 30% price cut on a $100M base isn’t a rounding error — it’s tens of millions of dollars, which is exactly the kind of math that makes “the model is slightly better” a secondary argument.

This Isn’t an Isolated Account

The Information’s reporting didn’t stop at Replit. At Rippling, the HR software company, Anthropic models handled 57% of coding work across hundreds of engineers back in May. By October, that had flipped: OpenAI models now handle 52% of that work, Anthropic 18%. Rippling CEO Parker Conrad attributed the shift to cost — OpenAI got cheaper, and the company moved its workload to match.

Separately, the-decoder reported that DataCamp’s CEO said switching to OpenAI’s pricing cut the company’s model costs by 80 to 90%. That’s not a 30% discount story anymore — that’s a different cost structure entirely, at least for DataCamp’s specific workload.

Three companies, three separate deals, one pattern: Anthropic’s enterprise pricing is getting undercut hard enough that cost, not capability, is the deciding variable. This site covered Anthropic overtaking OpenAI in U.S. business spend back in May, on the strength of Claude Code specifically. We flagged then that the lead was “built almost entirely on one product” and that the same product’s economics were “forcing budget conversations” even among the customers driving the number. Six months later, that’s not a hedge anymore. It’s the headline.

Why This Matters

Here’s the uncomfortable part for Anthropic: coding is the one category where Claude built its reputation this year. Losing share there on price, rather than on a benchmark, is a worse look than losing a generalist-chatbot comparison — because it means the thing keeping customers isn’t preference, it’s whichever vendor is willing to eat margin to keep the account. That’s a fragile kind of loyalty, and it cuts both ways. If OpenAI can buy Rippling and Replit’s business with a discount, nothing stops Anthropic from buying it back the same way next quarter, except whatever decision it already made about how much margin it’s willing to sacrifice to keep winning on spend share.

It also complicates the clean narrative either company would prefer. Anthropic’s public story is “best model for serious engineering work.” That story doesn’t address what happens when the better-performing-on-paper model costs 30-90% more for the same output and a competitor is actively subsidizing the switch. OpenAI’s story is “we’re winning on merit.” The reporting here says the merit argument is doing less work than the subsidy.

For buyers, the lesson isn’t “OpenAI is now better for coding” or “Anthropic lost.” It’s that enterprise AI pricing in late 2026 is a live negotiation, not a published rate card, and the vendor willing to discount harder is currently OpenAI — at least for high-volume coding accounts. That can flip. Our running pricing comparison tracks published tiers, but published tiers are increasingly the starting point for a negotiation most $100K+/year accounts are apparently having anyway.

What Are Your Options Now

If you’re running a high-volume coding workload on Claude, ask your account team directly whether a renewal discount is on the table before you assume it’s automatic. Replit’s experience suggests Anthropic is willing to let volume discounts lapse rather than extend them without a bigger commitment — which means the default outcome of doing nothing is a price increase, not a renewal.

If you’re evaluating OpenAI’s models for coding work you currently run on Claude, do the side-by-side on your own workload before switching defaults. Claude vs ChatGPT for coding still holds up as a capability comparison — the economics changing doesn’t mean the quality gap (in either direction) disappeared. Replit changed its default model provider; it didn’t claim Claude stopped working.

If you’re locked into a single-vendor agentic coding tool, this is a reminder that your tool’s default model is a business decision made on your behalf, for reasons that may have nothing to do with your workload. Check what model your coding agent is actually routing to by default, and whether switching it costs you anything beyond a settings change.

The Bigger Picture

Zoom out and this is the AI coding market doing what commodity markets do once the capability gap narrows enough that price becomes the tiebreaker. DeepSeek cut its own API pricing 75% earlier this year and opened a 6-7x cost gap with Claude and GPT — proof the undercutting pressure isn’t just a two-company fight. OpenAI discounting 30%+ to take Anthropic’s coding accounts is the same dynamic playing out between the two biggest labs instead of a challenger and the incumbents.

That’s a genuinely different competitive phase than the one this site spent most of 2026 covering, which was almost entirely about capability — which model reasons better, which one hallucinates less, which one you can trust with an agent’s permissions. Price wars don’t replace that conversation. They sit on top of it, and they move faster, because a CFO can approve a vendor switch in a budget cycle in a way that a capability gap takes a model generation to close.

Our Take

I don’t think Anthropic lost this fight because Claude got worse. The reporting doesn’t say that, and nothing else published this year supports it. Anthropic lost this specific negotiation because it made a margin decision — hold the line on discount terms, ask for more commitment instead of giving more discount — and OpenAI made the opposite bet on the same account. That’s a legitimate strategic choice for Anthropic to make. It’s also a choice with a visible cost now attached to it: a flagship coding customer, a CEO willing to say “ChatGPT is better now” on stage, and a pattern repeating at Rippling and DataCamp.

What I’d push back on is treating Masad’s Disrupt comments as an honest read on model quality. He said one thing on stage and a softer thing to TechCrunch within a day, and the actual reporting on the deal structure — discount offered, discount declined, spending reallocated — reads like a pricing story that got narrated as a product story because “ChatGPT is better now” is a better soundbite than “OpenAI’s finance team outbid Anthropic’s.” Watch what Anthropic does with its next few enterprise renewals before deciding whether this was a one-off or the start of a pattern it’s willing to live with.

Frequently Asked Questions

Did Replit stop using Claude entirely?

No. Replit’s documentation still lists seven Claude models, and users can select them manually in Replit Agent. What changed is the default: OpenAI’s models now power Replit Agent out of the box, where Anthropic’s previously did.

Why did Anthropic lose Replit’s business?

According to The Information, Anthropic declined to extend Replit’s existing pricing discount without Replit committing to higher spending. OpenAI offered a discount of more than 30% on a contract reported at $100 million-plus a year, and Replit moved most of its spend there instead.

Is this just about Replit, or is it a bigger trend?

It’s bigger than one account. The same Information report describes Rippling’s coding-work share flipping from 57% Anthropic / 16% OpenAI in May to 18% Anthropic / 52% OpenAI by October, driven by cost. The Decoder reported DataCamp’s CEO saw an 80-90% cost reduction after switching. Three separate companies cited price, not performance, as the deciding factor.

What did Replit’s CEO actually say?

At TechCrunch Disrupt on October 16, Amjad Masad said Replit was no longer working with Anthropic and that “ChatGPT is better now,” citing cost, partnership, and performance. He walked that back within a day, saying Replit “remains model agnostic,” and a source close to the account told TechCrunch the Anthropic partnership was still active.

Does this mean Claude is falling behind GPT on coding tasks?

Not according to the reporting. Every account cited — Replit, Rippling, DataCamp — frames the switch as a pricing decision, not a capability verdict. Claude vs ChatGPT for coding remains a workload-by-workload call; nothing in this story changes which model writes better code, only which one costs less for high-volume use.

Should I expect similar discounts if I ask Anthropic or OpenAI?

Possibly, if you’re negotiating a contract in the range these companies are — six or seven figures annually. There’s no evidence either lab is extending 30%+ discounts to smaller accounts. Published pricing is still the realistic baseline for most individual and small-team use.

Will Anthropic match OpenAI’s discount to win Replit back?

Unknown, and nothing published so far says Anthropic plans to. The company’s decision to decline Replit’s discount renewal in the first place suggests it was willing to accept losing volume share on this account rather than lower its price further. Whether that holds after seeing the account move is a separate decision it hasn’t publicly made yet.

---

Last updated: October 25, 2026. Sources: The Information — Anthropic’s Margin Is OpenAI’s Opportunity · Techmeme · TechCrunch — Replit is no longer working with Anthropic, CEO says · The Decoder — OpenAI offers discounts up to 90 percent to poach Anthropic’s biggest customers.

Related reading: Replit Agent Review 2026 · Anthropic Just Beat OpenAI in Business Spending · Claude vs ChatGPT for Coding · AI Tool Pricing in 2026 · DeepSeek’s 75% Price Cut