OpenAI Weighs Drastic Token Price Cuts as Anthropic Eclipses Its Valuation and Captures Enterprise Coding
OpenAI is weighing significant reductions to its per-token API pricing, according to people familiar with the matter cited by the Wall Street Journal. The discussions are early and nothing is decided. But the strategic logic behind them is visible in the numbers that have accumulated since April.
This is not a cost-optimisation exercise. It is a response to a competitive shift.
What Changed
Three data points pushed OpenAI toward the pricing conversation:
Valuation reversal. Anthropic closed a $65 billion funding round in late May, pushing its valuation to roughly $965 billion — ahead of OpenAI’s $852 billion mark from March 2026. For the first time, Anthropic is valued higher. The S-1 narratives both companies are building toward public markets now require them to explain why.
Claude Code’s revenue inflection. Anthropic’s coding assistant crossed $1 billion in revenue within six months of its launch. That rate of growth is described internally at OpenAI as “mind-blowing.” On the Ramp AI Index, which tracks enterprise software spending, more companies are now paying for Anthropic than for OpenAI. That is a first.
ChatGPT traffic decline. ChatGPT’s share of global generative AI web traffic fell from 77.6% in May 2025 to 53.7% in April 2026 — a 22-point drop in 11 months. Claude’s traffic over the same period is up 306%.
The Price Landscape
Current frontier pricing sets the frame:
| Model | Input (per 1M tokens) | Output (per 1M tokens) |
|---|---|---|
| Claude Fable 5 | $10 | $50 |
| GPT-5.5 | $5 | $30 |
| Claude Opus 4.8 | $5 | $25 |
| DeepSeek V4-Pro | $0.14 | $0.87 |
GPT-5.5 is already cheaper than Fable 5 by a wide margin on paper. The problem for OpenAI is that enterprise developers are not choosing on list price alone — they are choosing on which tool their engineers actually want to use, and a large cohort has migrated to Claude Code. Price cuts would narrow the per-token gap but would not by themselves reverse that preference.
Sam Altman acknowledged costs as “a huge issue” for enterprise customers at the Intelligence at Work event, saying the company was exploring ways to “help people get more value for less spend.” That framing was public signal. The WSJ reporting suggests the internal discussions are more concrete than that phrasing implied.
The IPO Bind
Both companies filed confidential S-1s this week. The tension that creates is direct: cutting token prices compresses the revenue margins that public-market investors will scrutinise in those filings.
OpenAI posted an adjusted operating loss of $1.22 for every dollar of revenue in Q1 2026. Anthropic is approaching its first profitable quarter. A token price war initiated by the less-profitable company, in the weeks before both firms seek trillion-dollar public valuations, is not an obvious move. The discussions are happening anyway because the alternative — losing the enterprise coding market permanently — looks worse.
The precedent that hangs over both companies is DeepSeek V4-Pro, which locked in a permanent 75% price cut in June, putting frontier-quality output at $0.87 per million tokens. Google has been cutting Gemini API prices throughout 2026. The commodity floor is rising faster than either US incumbent planned for.
What Developers Would Get
A meaningful cut to GPT-5.5 token prices would shift the economics for the segment of enterprise buyers still running OpenAI workloads who are on a per-token budget. Uber set usage caps on AI tools after burning through its annual AI budget four months early. Walmart similarly limited its internal agent. Both are symptoms of the tokenmaxxing problem — organisations that approved budgets at 2025 prices for 2026 usage patterns.
The developers who have already moved to Claude Code are unlikely to reverse based on pricing alone. The developers still on OpenAI, running cost-sensitive batch workloads, would benefit immediately.
Whether the cuts materialise, and at what magnitude, is still unclear. The WSJ’s people-familiar-with-the-matter sourcing puts this firmly in the “directional signal, not announcement” category. What is clear is that the competitive dynamics that created this conversation are not going to reverse on their own.