OpenAI Eyes Token Price Cuts as Anthropic Overtakes It

OpenAI is weighing major token price cuts, just days after the company confidentially filed for an initial public offering, according to a report on OpenAI’s pricing review published this week. The planned reductions, which would lower what developers and enterprises pay to run OpenAI’s models, come as Anthropic has pulled ahead in both valuation and the enterprise market OpenAI once owned outright.

OpenAI is preparing to go public “as soon as this fall,” according to the report, though the company has said it has not decided on timing. Anthropic, the five-year-old rival behind the Claude family of models, filed its own confidential S-1 with the Securities and Exchange Commission on June 1, ten days before OpenAI’s filing. Both companies are now public-market bound within weeks of each other, and both face the same question: can a market in which buyers are actively balking at AI bills sustain the kind of growth that justifies the valuations they have been raising at?

Anthropic Closed the Gap, Then Pulled Ahead

Anthropic’s Series H round in late May set the company’s post-money valuation at $965 billion, with a fresh $65 billion raised. OpenAI’s most recent private valuation, set in March after a $122 billion fundraise, sat at $852 billion. The flip was the first time Anthropic had topped its larger rival on private-market value, and the timing matched a parallel lead in revenue.

Anthropic told the AP this week that it is now generating $47 billion in annualized revenue from selling access to its Claude models. A separate write-up of the same fundraise said the company was on track to hit $50 billion within a month and that its annualized revenue had grown 80-fold in the first quarter. The growth has been driven in large part by Claude Code, Anthropic’s agentic coding tool, which has become the default assistant for a widening share of professional software engineers and the wedge that has reopened the enterprise market for Anthropic. Anthropic was formed in 2021 by ex-OpenAI leaders, and its rise has rewritten the assumption that the company that released ChatGPT first would keep the commercial lead indefinitely.

OpenAI, by contrast, has spent the spring walking back internal targets. A separate report in April said the company had missed several monthly sales goals for 2026 and fallen short of a 1 billion weekly active users milestone, with Anthropic named as the source of the share loss in coding and enterprise. Patrick Corrigan, a law professor at Notre Dame University who studies IPOs, told the AP that the back-to-back filings are “a little bit surprising,” since “public investors are going to be comparing them roughly around the same time.” Wedbush Securities analyst Dan Ives called the Anthropic filing “an opening of the floodgates for the IPO market, which has been relatively dormant for a few years.”

The Token Meter Is Breaking Enterprise Budgets

The price-cut rethink did not start with the IPO race. It started with the bills. The practice now widely called tokenmaxxing, pushing models to consume as many tokens as possible on coding, research, and customer-service tasks, often without clear returns on the spend, has been building for the better part of a year, and the cracks are showing in public.

Uber chief technology officer Praveen Neppalli Naga told The Information that his 2026 AI budget, set at the start of the year, was gone by April. “I’m back to the drawing board because the budget I thought I would need is blown away,” Naga said. The driver was Claude Code: according to a Fortune report cited in a recent look at runaway Claude invoices, the share of Uber engineers using the tool climbed from roughly 32% in February to between 84% and 95% by April. Uber’s AI cost trajectory was the same one the rest of the industry has been running on, and Claude Code sat at the top of the list. The result was a finance team rewriting its 2026 plan in the second quarter. The company’s research-and-development line rose 17% year on year to $951 million in the first quarter.

Microsoft, the most public adopter on the corporate side, told employees in its Experiences and Devices division this month to wind down most of their internal Claude Code licenses by June 30, 2026, and to shift work to GitHub Copilot CLI. Engineers there had been posting per-head monthly bills between $500 and $2,000. The decision was a procurement call, not a quality call: Microsoft owns the competing product, the competing product runs on its own infrastructure, and every dollar paid to Anthropic is a dollar that could route to a Microsoft cost center instead.

The extreme case is an unnamed enterprise that ran a roughly $500 million Claude bill in a single month after deploying licenses widely without per-employee token caps, an AI consultant told Axios. The consultant did not identify the client. Anthropic’s published rate card makes the math legible: Opus 4.7 lists at $5 per million input tokens and $25 per million output tokens, Sonnet 4.6 at $3 and $15. Opus 4.8, covered in our recent coding benchmark review, is the model most enterprise customers have been running Claude Code on for the past six months. A single agentic coding session that reads a repository, plans an edit, and iterates can burn millions of tokens in an afternoon. Three different price tags, one shared mechanic: the meter rewards the same behaviour the productivity pitch promised, and the bill shows up before the productivity line does.

Altman Calls Cost “A Huge Issue”

OpenAI chief executive Sam Altman acknowledged the pressure directly at a Tuesday enterprise event earlier this month, in remarks covered by a write-up of the company’s enterprise gathering. Altman told the audience that cost had gone from a non-issue at the start of the year to a “huge issue” almost overnight, citing customer memes of the form “My company spent my entire 2026 budget in Q1.” The event was the same week OpenAI was finalizing its confidential S-1.

The framing is notable for two reasons. First, Altman’s own company benefits when customers spend more on tokens, so an open acknowledgment of sticker shock from the seller’s podium is unusual. Second, the line puts OpenAI’s CEO on record aligning with the cost-control case that Anthropic’s enterprise wins have been making by example. The pricing review published this week said OpenAI is “evaluating reductions” in the prices it charges for tokens and expects Anthropic to move in the same direction. The defensive posture, written into a WSJ scoop from inside the company, is itself the story.

That went from, at the beginning of this year, an issue that never came up, people were totally happy with the amount they were spending, to all of a sudden, a huge issue.

Altman added that OpenAI would help customers “get more value for less spend.” The phrase is a pricing signal in plain English. The same coverage of the enterprise event quotes a Google software engineer putting the working diagnosis bluntly on X: “getting value out of agents is still too difficult for most engineers, so they end up just burning tokens.”

What a Token Price War Could Look Like

The pricing review is short on numbers and long on direction. OpenAI is “evaluating significant reductions,” “anticipating that Anthropic will make similar moves,” and acting from a position of having “been trying to catch up with Anthropic in winning enterprise customers.” The shape of the cut will only become public when OpenAI files its S-1, which is what makes the IPO itself a pressure point for the move.

That timing is the whole story: a token price cut announced just before a public listing is a margin haircut that institutional buyers will price in, while a token cut announced after a successful offering is a margin haircut the company can absorb. A second-mover cut would be a different kind of signal, an acknowledgment that the first move worked and that the market has already settled on a new price. Both companies are now calibrating their S-1 disclosures against that possibility.

For buyers, the practical effect depends on which side moves first. The two companies now sell near-substitutable models for the most common enterprise workloads, and switching costs between OpenAI and Anthropic are low. The same coverage called interchangeability the structural feature that would make any first-mover cut force a rapid response. The same dynamic put Anthropic’s lower-priced tiers in direct competition with OpenAI’s own discount models, and is now pulling the headline-priced Opus-class and Fable-class products into a price race. Fable 5, the public version of Anthropic’s Mythos-class model, launched on June 9 and is covered in our review of the new model.

Three procurement resets are already visible across the buyer side, and they are the same levers that will let a price cut reach the bottom line of an enterprise invoice:

• Per-seat token ceilings. Hard monthly allowances per engineer, with overage requiring manager approval. The $500 million Axios invoice is the textbook case for this control.
• Internal substitutes. Microsoft is rerouting Claude Code spend back to GitHub Copilot CLI. Companies with their own model infrastructure are doing the same.
• Cheaper-model defaulting. Routing routine work to Haiku 4.5 at $1 input and $5 output per million tokens, and reserving Opus for genuinely hard tasks, cuts the headline rate by 80 percent without removing the tool.

For OpenAI, the calculus also includes a balance-sheet risk the April reporting flagged. The same coverage that surfaced the missed internal sales targets said the company has been raising at scale to keep up with the compute required to serve a growing user base. Cheaper tokens pull the per-user revenue line down at the moment when the public-market story needs it to keep climbing.

Frequently Asked Questions

When will OpenAI’s token price cuts take effect?

The pricing review reported this week says OpenAI is “evaluating” significant reductions. The company has not announced specific prices or a date, and the same reporting noted that OpenAI is acting from a position of trying to catch up to Anthropic in the enterprise market. Cuts are likely to land before or alongside the public listing OpenAI says could happen as soon as this fall.

Is this an AI price war?

Not yet, in name. The pricing review reported that OpenAI is “anticipating that Anthropic will make similar moves” if it cuts, which is the standard setup for a price war in any commodity market. The first public cut by either company will be the moment the language changes.

What does “token” pricing mean for buyers?

Tokens are the units AI companies use to meter usage. A simple chat exchange might consume a few thousand tokens; an agentic coding session that reads a repository and writes a fix can consume millions in a single afternoon. Pricing is set per million tokens, split between input (the prompt and any context the model reads) and output (the model’s reply). The total bill scales with workflow, not with seat count.

Why is Anthropic pressing OpenAI now?

Two reasons meet at the same week. Anthropic closed a $65 billion Series H that valued it at $965 billion on May 28, then filed a confidential S-1 with the SEC on June 1, taking the IPO lead. OpenAI filed its own confidential S-1 on June 8, with a private valuation of $852 billion. Anthropic’s enterprise coding share, driven by Claude Code, has been the wedge that closed the valuation gap, and it is the same wedge now forcing OpenAI to defend with price.

How big is the gap between the two companies?

On the most recently reported numbers, Anthropic’s private valuation is $113 billion higher than OpenAI’s ($965 billion against $852 billion) and its annualized revenue, per the AP, stands at $47 billion. OpenAI’s larger user base and earlier-mover position in consumer AI are not in the public filings yet, and the S-1 disclosures later this year will be the first detailed look at the side-by-side financials.

Figures and product names are accurate as of June 11, 2026. Pricing for AI tokens is set by each provider and is subject to change.