IndustryNews

Replit shifts AI spend from Anthropic to OpenAI, as reported: what Replit has said and what's unconfirmed

Replit switched AI spending from Anthropic to OpenAI after OpenAI offered more than 30% off a contract worth over $100 million a year, The Information reported on October 24, 2026, according to The Decoder. What Replit's CEO has said on the record, what Replit's docs show, and what remains unconfirmed.

A long glass price bar with its last 30% lifted out as a lit piece over a dashed gap, for OpenAI's reported discount to Replit, on fuchsia

What’s new, in 20 seconds

  1. The Information reported on October 24, 2026, that Replit switched AI spending from Anthropic to OpenAI after OpenAI offered more than 30% off a contract worth more than $100 million a year, according to The Decoder's summary of the paywalled story. Aggregator summaries by Techmeme and RuntimeWire say it was most of Replit's AI spend. Replit, OpenAI and Anthropic haven't confirmed the terms.
  2. Replit CEO Amjad Masad said at TechCrunch Disrupt, as TechCrunch reported on October 16 and updated October 17, that OpenAI's models are the default for Replit Agent, a choice 'based on both cost, partnership, and performance.' He later said 'Replit remains model agnostic.'
  3. Replit's model-selector docs, checked October 26, 2026, still list Claude Opus 5.5, Claude Sonnet 5 and other Claude models alongside GPT-6 Sol and GPT-6 Astra. Masad told TechCrunch Replit had been one of Anthropic's top five customers.
Contents

Replit, the AI app-building and coding agent platform, switched AI spending from Anthropic to OpenAI after OpenAI offered a discount of more than 30% on a contract worth more than $100 million a year, The Information reported on October 24, 2026, according to The Decoder's summary of the paywalled story, headlined "Anthropic's margin, OpenAI's opportunity." Summaries by two aggregators, Techmeme and RuntimeWire, say it was most of Replit's AI spend. Replit's CEO had already said on the record that OpenAI's models are the default for Replit Agent. Anthropic's costs and customer concentration, which our report on its draft IPO prospectus covers, are the backdrop.

This story is based on TechCrunch's reporting from TechCrunch Disrupt, OpenAI's August 19 announcement with Replit, Fortune's report the same day, Replit's documentation, and The Information's report as relayed by The Decoder and two aggregators, RuntimeWire and Techmeme, all checked October 26, 2026.

The short version

  • Reported (The Information, via The Decoder): Replit switched to OpenAI under a contract worth more than $100 million a year with more than 30% off. The aggregators' summaries add that it was most of Replit's AI spend.
  • On the record (Masad, via TechCrunch): OpenAI is the default and "preferred partner" for Replit Agent, chosen on "cost, partnership, and performance," but "Replit remains model agnostic."
  • In Replit's docs: Claude models remain selectable, next to OpenAI's GPT-6 models and Kimi K3.
  • Unconfirmed: the dollar value, the discount, and whether Anthropic ended a discount of its own.

What did The Information report?

The Information's article is paywalled, so 9to5AI relies on outlets that summarized it:

DetailRelayed byNote
Replit shifted most of its AI spending from Anthropic to OpenAI in recent monthsRuntimeWire (October 25); Techmeme listing (October 24)Aggregators only; RuntimeWire attributes it to "a person familiar with the situation." The Decoder says only that Replit "switched"
OpenAI's discount was more than 30%The Decoder (October 25); RuntimeWire; TechmemeRuntimeWire says the discount is on model prices
The contract is worth more than $100 million a yearThe Decoder; RuntimeWire; Techmeme—
The switch followed Anthropic discontinuing a discount it had given ReplitTechmeme's headline for the story onlyNot in The Decoder's or RuntimeWire's summaries

Table 1The discount and contract value appear in The Decoder's summary as well as the aggregators'; the share of spend and Anthropic's ended discount appear only in aggregator summaries. None has been confirmed by Replit, OpenAI or Anthropic. Source: The Decoder, RuntimeWire, Techmeme, checked October 26, 2026.

Last verified: October 26, 2026

The Decoder, summarizing the same report, said Replit is one of several customers OpenAI has won with aggressive pricing. It relayed Rippling CEO Parker Conrad's figures that OpenAI's share of Rippling's coding workload rose from 16% to 52% while Anthropic's fell from 57% to 18%, and DataCamp CEO Jonathan Cornelissen's statement that OpenAI's pricing cut its costs by 80% to 90%. It also said OpenAI is partly buying market share at the expense of its own margins.

What has Replit said?

Masad spoke onstage at TechCrunch Disrupt, and TechCrunch first published his comments on October 16, 2026, under a headline saying Replit was no longer working with Anthropic. TechCrunch later updated the story and changed the headline to "Replit makes OpenAI its default over Anthropic, CEO says."

According to TechCrunch:

  • Masad said the decision was "based on both cost, partnership, and performance," and that "I think ChatGPT is better now." He said OpenAI's Sol and Astra models "perform better on building vibe-coded applications from scratch," citing an internal benchmark and a public one he said Replit publishes. TechCrunch didn't name either benchmark or report scores.
  • He clarified that customers can still choose Anthropic models but that OpenAI is the default, and TechCrunch noted most customers keep the default, which could sharply cut Replit's Anthropic spending.
  • In a statement, he said "Replit remains model agnostic" and that OpenAI is the "preferred partner" for Replit Agent, "based on performance, economics, and the strength of our partnership."
  • He said Replit had been one of Anthropic's top five customers until the change.
  • An unnamed insider disputed his account and said the Anthropic partnership was still in place, TechCrunch reported. TechCrunch also pointed to a late-September Replit blog post citing Claude Fable 5 and 5.1 in Replit Agent's production environment. That September 29 engineering post compares Fable 5, Fable 5.1 and OpenAI's GPT-6 Astra, and says Replit evaluated Agent's Max mode "with Astra as the core loop."

Replit's own model-selector documentation, checked October 26, 2026, lists both companies' models. Power Mode offers GPT-6 Sol, GPT-6 Luna Fast, Claude Sonnet 5 and Claude Sonnet 4.6; Max Mode offers GPT-6 Astra, Claude Fable 5.1, Claude Opus 5.5, Claude Opus 4.8, Claude Opus 5, Claude Opus 5 Fast and Kimi K3. When the product shows "Auto," Replit routes each task to a model itself, and Free Mode always uses that routing. The docs carry no date and say availability "can vary by rollout, organization settings, and authorization." For what each OpenAI model is for, see our GPT-6 Astra, Sol and Luna guide.

How did Replit's OpenAI partnership start?

It was public before the spending report. On August 19, 2026, OpenAI announced that Replit's new Free Mode runs on GPT-5.6 Luna, with harder tasks routed to GPT-5.6 Sol. Masad said in OpenAI's post that the model made it possible "for us to offer it to millions of users in Free Mode."

Fortune reported the same day that Free Mode still requires a paid Core ($20 a month) or Pro ($100 a month) plan, that it is the default for those users, and that OpenAI's 80% price cut for Luna on July 30 made it possible. The Fortune report didn't mention Anthropic.

Masad also told TechCrunch that enterprise customers were about 2% of Replit's revenue in 2025 and about 15% in 2026, and that more than half of revenue is business-to-business. He declined to share annual recurring revenue.

Why does this matter for Anthropic?

By Masad's account, Replit had been one of Anthropic's five largest customers, so losing default status there shifts real volume. Per the Wall Street Journal, as relayed in our prospectus story, about 20% of Anthropic's first-half 2026 revenue came from two customers, and Anthropic spent nearly $9 billion on compute in the half against $16.3 billion of revenue. Those reports didn't name Replit, and its size relative to those customers isn't public.

For OpenAI, the reported deal fits a pattern of discounting to win share, which The Decoder says comes partly at the cost of margin. None of the three companies has confirmed the reported contract terms. Replit had a prominent agent incident in July 2025, when its agent deleted a user's production database, as our report on unattended agents recounts; that isn't connected to the provider switch.

What's confirmed, reported and uncertain

  • Confirmed: OpenAI's August 19 Free Mode announcement with Replit; both providers' models in Replit's docs; Replit's September 29 post evaluating Agent with GPT-6 Astra as the core loop.
  • On the record but secondhand: Masad's statements, as reported by TechCrunch.
  • Reported (The Information, via The Decoder and aggregators): the switch, the more-than-30% discount, the $100 million-plus annual value.
  • Aggregator summaries only: that it was most of Replit's AI spend (Techmeme, RuntimeWire); that Anthropic had discontinued a discount for Replit (Techmeme's headline alone).
  • Uncertain: how much Replit still spends with Anthropic; whether the Anthropic partnership continues in any formal sense; Anthropic's and OpenAI's comments.

Read next

Comments (0)

More on Coding agents

View all
Developers · NewsCursor Computer Use: Cursor's agent can now drive desktop apps and the iOS SimulatorCursor Computer Use, announced October 23, 2026, lets Cursor's agents click and type in desktop apps such as the iOS Simulator to test what they build. What Cursor has confirmed, what its docs say about permissions, and what it hasn't said yet.·5 min read
Agents · NewsNVIDIA OpenShell: the open-source runtime that sandboxes fleets of AI agents, and what it actually enforcesNVIDIA OpenShell is an Apache 2.0 runtime that runs AI agents such as Claude Code and Codex in sandboxes, checks every network request against a policy and keeps real credentials out of the agent's reach. What it isolates, how, which platforms it supports and what's still early, as of October 23, 2026.·6 min read
Developers · GuideLiteLLM 1.107.0 PyPI compromise: which versions and Docker tags were affected, how to check, and what to rotateThe LiteLLM 1.107.0 PyPI compromise put a credential stealer and a GitHub-driven backdoor into every 1.107.0 wheel and four Docker tags on October 21, 2026. Which versions are safe, how to check your machines, and what LiteLLM says to rotate.·8 min read
Open source · NewsRecursive Flint: an Apache 2.0, 40B coding model with 2.5B active parameters and a self-reported 89.7% on SWE-bench Pro v2Recursive Flint is a 40B-parameter mixture-of-experts coding model with about 2.5B active parameters, released under Apache 2.0 on October 21, 2026. Its 89.7% on SWE-bench Pro v2 is Recursive's own number and wasn't on Scale AI's leaderboard as of October 22.·5 min read
The Week in AI · every Friday

Get the cluster, not just the headline.

The week’s biggest stories in AI and tech, what they mean, and what to read next.

Free. One email a week. Unsubscribe in one click.