Replit for Business Is Live: AI Agents for Every Team, Backed by a $100M OpenAI Switch

The short version: Replit launched Replit for Business on October 26, 2026, a version of its AI agent platform built for whole companies rather than individual builders. It adds the controls IT departments have been asking for since coding agents went mainstream: orchestration across agents, sandboxed execution that keeps agents away from production systems, identity controls, and the option to run agents inside a company's own cloud accounts, with SOC 2 Type I, ISO 27001 and ISO 42001 certifications listed on Replit's enterprise page. It arrives three weeks after Replit's chief executive said the company had moved its default model from Anthropic to OpenAI — a switch The Information reported on October 24 was driven by a discount of more than 30% on a contract worth over $100 million a year.

That timing is not a coincidence. Replit is trying to turn a developer tool into enterprise infrastructure, and the two announcements are the two halves of the same pitch: the product now has the governance a CIO needs, and the model behind it now comes from the vendor willing to discount hardest. Whether the combination is actually ready for production is a fairer question than either announcement answers on its own, and the pricing is where the honest caveats live.

What Replit for Business actually is

Replit started in 2016 as a browser-based coding environment. Its first AI agent shipped in 2024, and Agent 3 followed in September 2025 — an agent that builds other agents, powered by the Mastra framework, which Replit says spins up thousands of isolated sandboxes a week and reaches 90% autonomy and 96% workflow success. Those figures are Replit's own, reported by Mastra, and no independent test has verified them. But the direction is clear: Replit's agents stopped being a coding assistant and became a way to build software and automations without writing code at all.

Replit for Business is the enterprise wrapper around that shift. According to the launch coverage, it offers orchestration of agents across a business, sandboxing that isolates agents from production systems, and identity controls, while maintaining SOC 2 Type I compliance. Corporate teams can run Replit tools within their own cloud accounts, which matters for data control and residency — the two objections that kill most enterprise AI pilots before they start.

Replit frames the product with a phrase of its own: the "Self-Driving Company." In the launch announcement, the company argues that agents will change how companies run in the same way the PC and the cloud changed computing before them — people set goals and stay accountable, agents take on the execution, and every team gains the ability to pursue work that once needed more people or more skills. The company says its own engineering teams nearly tripled their code output in the first half of 2026 while review times and product incidents stayed flat. That is Replit's own account of its own teams, not an independent measurement, but it is the number the company is staking its pitch on. One naming note: the launch announcement calls it "Replit for Business," while Replit's own site now presents the enterprise tier as Replit Enterprise — the replit.com/business address returns a 404, so "for Business" is the launch name and "Enterprise" is the live product page.

Replit's enterprise page fills in the governance picture in its own words. Agents "build, test, and investigate in isolated cloud environments, away from employee laptops," with required security scans and engineering review before anything ships. Access is managed with SSO, SCIM and role-based permissions, and connector credentials are kept outside the agent's context. The "bring your own cloud" option runs the data plane in the company's own cloud tenant. The page lists SOC 2 Type I, ISO 27001 and ISO 42001 certifications, searchable audit logs, and 510+ connectors and custom integrations. It also claims 86% of the Fortune 500 use Replit — Replit's own figure, not an independent count.

The commercial backdrop helps explain the urgency. Enterprise contracts accounted for 2% of Replit's revenue in 2025 and have grown to 15% in 2026, according to launch coverage. The company raised a $250 million Series C at a $3 billion valuation in September 2025, when it reported $150 million in annual revenue, and reporting puts a $400 million Series D at a $9 billion valuation in March 2026. Replit for Business is the product that justifies that trajectory — or the one that has to, if the enterprise push stalls.

The OpenAI switch: what happened, in order

The model switch is the part of this story with named sources and numbers, and it is worth laying out in sequence because the reporting has moved in stages.

October 17: Replit chief executive Amjad Masad said Replit was no longer working with Anthropic and that OpenAI was now the default model for Replit Agent, citing cost, partnership and performance, and saying ChatGPT was better now. He later softened that position, describing OpenAI as the preferred partner for powering Replit Agent, while a person familiar with the relationship disputed that the split was complete.

October 24: The Information reported the price of the move: OpenAI offered more than 30% off an annual contract worth over $100 million, after Anthropic declined to extend a discount it had been giving Replit unless the company committed to spending more. Seven Claude models remain available for customers who select them, according to reporting on the deal — OpenAI is the default, not the only option.

October 25–26: The wider picture filled in. At Rippling, chief executive Parker Conrad said Anthropic's models handled 57% of the company's coding work in May while OpenAI handled 16%; those shares have since flipped to OpenAI at 52% and Anthropic at 18%, and Conrad said the switch was about cost. DataCamp chief executive Jonathan Cornelissen said OpenAI's pricing cut his company's costs by 80 to 90%. A large American bank doubled its ChatGPT allocation to 500 employees at no extra charge, paying under $7 per user against a standard rate of $20 or more. An investor memo cited in the reporting indicated Anthropic's growth slowed from the second quarter to the third quarter, weeks ahead of its expected November public listing.

Two things matter about how this is reported. First, these are reported discount terms and customer statements, not published price lists — neither lab publishes an enterprise price book, and the exact terms of Replit's contract are not public. Second, no benchmark appears anywhere in the story. Replit builds a coding product, the use case where model differences are supposed to be most visible, and the reported trigger for the switch is a withdrawn discount, not a test result. That says something real about the market: at the application layer, wrapped in a harness with retries and scaffolding, the two models were close enough that a 30% price difference decided it.

The pattern fits the broader price war we covered in our analysis of OpenAI and Anthropic cutting AI prices by up to 80% in four months: public model prices have collapsed, and the private enterprise discounts are now doing the same work behind closed doors. Replit is the clearest named example of a customer moving on price alone.

What it costs: the four plans, and the credit math that matters more

Replit's pricing has two layers, and the subscription is only the first. Every plan includes a monthly credit allowance, and everything beyond that allowance — Agent work, publishing, databases, storage — is charged as pay-as-you-go usage under Replit's effort-based pricing model. A credit is worth exactly one dollar at the entry level, which makes Replit unusual: Core costs $20 and includes $20 of monthly credits, so the subscription is best understood as prepaying for usage with the workspace thrown in free. Prices are subject to tax depending on location, according to Replit's pricing page.

The fine print is where the bill actually lives. Replit charges per checkpoint based on how much work the Agent did — as little as $0.06 for a simple request, and multiple dollars for a large one. Failed operations still consume credits. Bulk credit purchases improve the rate: 100 credits for $100, 250 for $225, 500 for $440, 1,000 for $850 and 2,500 for $2,050, which is a dollar a credit at the bottom and 82 cents at the top. Core credits expire each month with no rollover; Pro credits roll over for one month.

Publishing is billed separately from the subscription. Autoscale hosting is $2 a month plus $0.60 per million compute units and $0.40 per million requests. Reserved VMs run $15 to $130 a month. Static hosting is free, with $0.05 per GB of transfer. Reviewers who have tracked real spending report that active builders spend 3 to 5 times their base plan in Agent credit overages — a range worth treating as a warning, not a guarantee, since it depends entirely on how much you build.

One setting deserves a line of its own: spending caps are disabled by default. Set them before any serious use. That is the single most practical piece of advice in this article, and it comes from reviewers, not from Replit's marketing.

A note on the numbers themselves: several pricing guides still quote Core at $25 a month, a price Replit cut to $20 on February 24, 2026, when it also launched the Pro plan and began sunsetting the Teams plan. The figures in this article were verified by a live read of Replit's pricing page on October 26, 2026, and against independent pricing guides updated in October 2026, and the $20 monthly / $18 annual figures are the current ones. Replit for Business itself has no published price — Enterprise remains quote-only, which is standard for the category but means the headline cost of the new product is whatever your sales conversation produces.

Replit for Business vs the field

Replit for Business is not the only way to put agents inside a company, and the honest comparison is against two very different kinds of rival: other vibe-coding platforms, and governed internal-tool platforms built for IT from the start.

The table's real story is the trade-off. Superblocks starts at $100 per seat per month — seat-based, predictable, no credit metering — and is built for governed internal tools on private data. Replit is credit-metered and usage-based, which means the cost varies with how much you build, but it covers a much broader range of apps and prototypes. The vibe-coding rivals are cheaper and simpler, but they are not trying to be enterprise infrastructure. Replit for Business is Replit's attempt to be both: the builder anyone can use, and the platform IT will sign off on.

Distribution is part of the bet too. Accenture has announced work with Replit on secure enterprise adoption, which puts Replit in front of the global services firms that actually sell software into large companies. Gartner projects 80% of government bodies will implement AI agents by 2028, according to launch coverage — the market Replit is aiming at is not hypothetical, but it is also not yet buying at scale. The same agent wave is reaching consumers directly — see our roundup of 7 AI agents you can text in 2026.

The catches

The price of the new product is unpublished. Replit for Business is quote-only. Every number in this article's pricing table belongs to the existing plans; the cost of the Business tier itself is whatever a sales conversation produces, and that is where the discount leverage in the OpenAI deal may or may not flow through to customers.

The discount story is reported, not published. The more-than-30% figure and the over-$100-million contract value come from The Information's reporting, corroborated by customer statements from Rippling and DataCamp. Neither OpenAI nor Anthropic has published enterprise price terms, and Replit has not confirmed the contract value. Treat the numbers as reported, not as a price list.

The Anthropic split may not be total. Masad's original statement said Replit was no longer working with Anthropic; he later softened it to OpenAI as the preferred partner, and a person familiar with the relationship disputed that the split was complete. Seven Claude models remain available in Replit's documentation. If your team has standardized on Claude, verify availability yourself before assuming it is gone.

Effort-based pricing punishes iteration. The same prompt costs different amounts depending on how much work the Agent does, and failed operations still consume credits. A team learning to use agents will burn credits on mistakes — which is exactly when spending caps matter most, and they are off by default.

The autonomy numbers are the vendor's. Agent 3's 90% autonomy and 96% workflow success, and the claim that Replit's own engineering output nearly tripled, are company-reported. No independent evaluation exists yet. The honest test is a pilot with your own workflows, measured against your own baselines.

Who should try it, and who should wait

Try it now if you are a company where non-technical teams already build with Replit, and IT's objection has been governance rather than capability. The sandboxing, identity controls and in-cloud deployment address the specific reasons those teams get told no. Start with a pilot, set spending caps on day one, and measure cost per delivered workflow rather than cost per seat — the seat price is not the bill.

Wait if you need predictable per-seat pricing, or if your agents will touch regulated data. Superblocks-style seat-based pricing is easier to budget, and Replit's credit model is the wrong shape for a finance team that wants a fixed number. Wait too if your team has standardized on Claude: confirm which models are actually available in the Business tier before committing.

For everyone else, the free Starter plan and the $20 Core plan are the honest way in. Build one real internal tool, watch what it costs in credits, and decide with your own numbers. That is also the test the whole "Self-Driving Company" pitch has to pass: not whether agents can build software, but whether they can do it at a price and a reliability level a business can plan around. Replit's own numbers say yes. The market, starting this week, gets to check.

FAQ

What is Replit for Business?
It is Replit's enterprise tier, launched October 26, 2026. It adds orchestration, sandboxed agent execution, identity controls and in-cloud deployment to Replit Agent, aimed at companies that want non-technical teams to build with AI agents under IT governance.

How much does Replit for Business cost?
Replit has not published a price; Enterprise is custom and quote-only. The underlying plans are public: Starter is free, Core is $20 a month ($18 billed annually) with $20 of credits, and Pro is $100 a month ($90 billed annually) with $100 of credits. Usage beyond the allowance is billed per credit under effort-based pricing.

Did Replit really drop Anthropic for OpenAI?
OpenAI is now the default model provider for Replit Agent, and Replit's CEO said in mid-October that the company was no longer working with Anthropic. He later softened that to OpenAI as the preferred partner, and seven Claude models remain available for customers who select them. The Information reported the switch followed a discount of more than 30% on a contract worth over $100 million a year.

What is a Replit credit worth?
One dollar at the entry level: Core costs $20 and includes $20 of monthly credits. Bulk purchases improve the rate to 82 cents per credit at the 2,500-credit tier. Credits cover Agent work, publishing, databases and storage, and Core credits expire monthly with no rollover.

Is Replit for Business safe for production systems?
It is designed to be: agents run in sandboxes isolated from production, with identity controls and SOC 2 compliance, and teams can run agents inside their own cloud accounts. That is the vendor's design claim, not an independent audit — a pilot with your own systems is the way to verify it.

Key takeaways

• Replit for Business launched October 26, 2026, adding orchestration, sandboxing, identity controls and in-cloud deployment to Replit Agent for whole companies.
• OpenAI is now the default model for Replit Agent after a reported discount of more than 30% on a contract worth over $100 million a year, with seven Claude models still selectable.
• Enterprise contracts grew from 2% of Replit's revenue in 2025 to 15% in 2026, and the Business tier is the product built to keep that growth going.
• Pricing is usage-based: Core is $20 a month with $20 of credits, Pro is $100 with $100 of credits, and the Business tier itself is quote-only with no published price.
• Spending caps are disabled by default and failed Agent operations still consume credits, so set caps before any serious use and measure cost per delivered workflow.

Replit's bet, as of October 27, 2026, is that the company that taught the world to build software in a browser tab can now teach whole companies to run on agents — and that price, not capability, decides the enterprise deal. The launch is real, the governance story is plausible, and the discount that paid for the model switch is reported, not published. The rest is a pilot away.

Sources: Replit's official LinkedIn launch announcement; live reads of replit.com/pricing and replit.com/enterprise on October 26, 2026; The Information's October 24 reporting via Silicon Report and AI Tools Recap; Wisevoter's October 26 launch report; Replit pricing guides re-verified October 2026 (Totalum, AI Alleyway); Mastra's Replit customer case; Superblocks' Replit Enterprise comparison. All vendor performance claims are flagged as company-reported in the text above.