OpenAI Reportedly Plans Drastic AI Token Price Cuts Amid IPO and Competitive Fire

OpenAI is weighing a major pricing about-face. On June 10, 2026, The Wall Street Journal reported that the ChatGPT maker is considering substantial reductions in the per-token fees it charges developers and businesses for using its models . The discussions are still preliminary, but they signal a new, more combative chapter in the AI market — one in which pricing power is shifting rapidly.

The potential for drastic token-price cuts represents an immediate and sharp reversal. Just seven weeks earlier, on April 23, OpenAI had doubled the price of its flagship model when it launched GPT-5.5 at $5 per million input tokens and $30 per million output tokens . That move itself represented a 4x increase from GPT-5's launch price in August 2025 . Companies that had built cost models around declining frontier-model costs were suddenly facing a very different pricing trajectory. Now, that trajectory may be bending downward again, faster than anyone expected.

The financial squeeze behind the pivot

OpenAI's reported willingness to cut prices is rooted in several converging financial pressures.

First, the company is no longer hitting its own internal goals. A late-April 2026 Wall Street Journal report noted that OpenAI fell short of recent revenue and new-user targets, stoking concern among some executives that the company might not be able to sustain its enormous data center spending . Chief Financial Officer Sarah Friar reportedly warned fellow leaders that the organization could struggle to finance upcoming computing agreements if revenue doesn't accelerate sufficiently .

Second, a long-standing unit-economics problem has become harder to ignore as the company prepares for public markets. A Reuters analysis from late 2025 found that less than 5% of users pay for the service — and even that paid revenue barely covers the immediate compute costs of generating responses, not to mention OpenAI's much larger overhead . The company's own numbers paint a stark picture: in 2025, OpenAI generated $13.1 billion in revenue but spent roughly $22 billion to do it, spending approximately $1.69 for every dollar earned . Gross margins fell from 40% in 2024 to just 33% in 2025, well below the company's own 46% target .

Third, the imminent IPO changes everything. After confidentially filing for an initial public offering that could value the company at up to $1 trillion, OpenAI can no longer rely on investors subsidizing below-cost pricing . Public-market investors will demand a path to cost recovery plus margin. Deutsche Bank and HSBC both project staggering cumulative losses: $143 billion to $207 billion through 2029 . Those projections make every pricing decision a high-stakes bet on whether market share can grow fast enough to offset the underlying cost structure.

A direct response to Anthropic's enterprise strength

The reported price discussions are explicitly framed as a competitive response to Anthropic. Multiple outlets carrying the WSJ report note that OpenAI is anticipating a "war for users" with its rival, which recently released its own Claude Fable 5 model .

Anthropic has quietly built a formidable enterprise position. According to the Menlo Ventures State of Generative AI report for Q4 2025 — widely cited as the most rigorous enterprise LLM spend survey available — Anthropic holds approximately 40% of the enterprise market for large language models, compared to OpenAI's 27% and Google's 21% . That 40% share gives Anthropic leverage that OpenAI, historically the consumer-facing brand, has not yet matched in the enterprise segment.

Both companies are simultaneously racing toward IPOs, which adds another layer of complexity to the competitive dynamic . Whichever company demonstrates stronger enterprise revenue growth and healthier unit economics in the months leading up to its public debut will be better positioned to command a premium valuation.

DeepSeek's price war reshapes the bottom of the market

The pressure isn't only coming from the premium end. DeepSeek, the Chinese AI lab, has been waging an aggressive price war that is fundamentally reshaping the cost floor for frontier-quality intelligence.

In late May 2026, DeepSeek made a permanent 75% price cut on its V4 Pro model, bringing input costs down to $0.27 per million tokens . That price point makes V4 Pro approximately 7x cheaper on inputs and 17x cheaper on outputs than Anthropic's Claude Sonnet or OpenAI's GPT-5.5 . Its smaller V4 Flash model undercuts Anthropic's Claude Haiku by a staggering 10x to 25x on a per-token basis .

Crucially, DeepSeek's models are not just cheap — they are also performant. V4 Pro achieves approximately 90.5% on the GPQA Diamond scientific reasoning benchmark, within a few percentage points of Anthropic's Claude Opus 4.8 and OpenAI's GPT-5.5 . The resulting price-to-capability gap is enormous: an 18.5x price difference for near-comparable capability on the benchmarks enterprise buyers evaluate most heavily .

DeepSeek's cost structure is the underlying driver. While the company's officially claimed training cost of $5.57 million is widely understood to exclude R&D, data labeling, and experimentation — the real figure is estimated at closer to $100 million — that is still roughly one-tenth of what OpenAI spends on a single training run . The cost-efficiency gap between the proprietary labs and the open-weight upstarts is real and growing.

The bifurcation of the AI model market

One of the most significant structural shifts in the AI industry over the past twelve months is the rapid disappearance of the middle ground. By April 2026, the market had split into two distinct economies: premium, proprietary, integrated stacks (OpenAI and Anthropic at higher price points) and ultra-cheap, open-weight alternatives (DeepSeek, and to a lesser extent xAI and Meta's Llama via API providers) .

The middle tier — models that offered solid performance at mid-range prices — has largely vanished . Developers and enterprises are increasingly forced to choose between paying a premium for the reliability, ecosystem integration, and enterprise-grade support of the top-tier providers, or cutting costs dramatically by adopting open-weight models that often match frontier performance on key benchmarks.

This bifurcation creates a dangerous strategic position for OpenAI. If it cuts prices to compete with Anthropic, it risks compressing the margins public investors will demand. If it holds prices high to protect profitability, it risks losing both enterprise share to Anthropic and cost-sensitive volume to DeepSeek. The reported discussions suggest the company may be tilting toward protecting share over protecting near-term margins.

What developers and enterprises should watch

Several indicators will signal which direction OpenAI's pricing actually takes — and what it means for the broader market.

The S-1 filing, when it becomes public, will be the most important document. Investors will scrutinize gross margins closely: anything below 50% signals sustained price pressure . API revenue as a percentage of total revenue will reveal whether OpenAI is prioritizing enterprise subscriptions over its developer ecosystem . Capex guidance will indicate whether the infrastructure spending that drives per-token costs is expected to rise or fall .

OpenAI has already begun shifting its infrastructure strategy ahead of the IPO, scaling back ambitious plans to build its own data centers in favor of leasing capacity from cloud partners like Oracle, Microsoft, and Amazon . That shift toward a more variable cost structure could give the company more flexibility to cut token prices without destroying margins — if the cloud partners pass through competitive unit economics.

For enterprise buyers, the short-term outlook is clear: per-token costs are likely to fall, and contracts should be negotiated with downward price-adjustment clauses. For developers, the question is whether OpenAI's price cuts — if they materialize — are deep enough to close the enormous gap with DeepSeek's alternatives. On current trajectories, they will not be. The AI pricing story of 2026 is not a single company's strategy. It is a structural repricing of intelligence itself, and every major lab is being forced to adapt.