OpenAI is considering sharp reductions to its token pricing in an effort to win users from Anthropic, the Wall Street Journal reported Wednesday evening. The company expects Anthropic to make similar cuts, according to sources familiar with the matter cited by CNBC.
The reported cuts go beyond OpenAI’s existing Flex Processing API, which already offers 50% discounts on standard pricing. The o3 model in Flex mode costs $5 per million input tokens versus $10 at standard rates, and $20 per million output tokens versus $40. The new reductions under consideration are reportedly “more aggressive” in scope, according to CNBC.
The Competitive Context
The timing is not subtle. OpenAI confidentially filed for an IPO with the SEC on Monday. Anthropic closed its Series H at a $965 billion valuation on May 28, slightly ahead of OpenAI’s $852 billion March valuation. Both companies are now racing to demonstrate user growth and market share before public listings.
OpenAI currently charges consumers $8, $20, and $100+ per month for access to GPT-5.5 models. Anthropic charges $17 per month for Claude Pro (annual subscription) and $100+ for Claude Max. On the consumer side, ChatGPT reached 1 billion monthly app users in May, hitting the milestone in roughly three years.
Google has already moved. Gemini model API prices have been cut multiple times, and Chinese providers like DeepSeek have reduced API pricing to what TradingKey describes as “nominal” levels.
What Token Price Compression Means for Agent Operators
The Wall Street Journal noted that enterprise customers can easily switch API providers due to high product interchangeability, according to CNBC’s reporting. That dynamic cuts both ways for teams running autonomous agents.
Autonomous agents consume tokens at volumes that dwarf interactive chat sessions. A single agent running persistent memory, tool calls, and multi-step reasoning can burn through millions of tokens per day. Price cuts at the API layer directly reduce the operating cost of these deployments. For agent frameworks like OpenClaw and Hermes that let operators swap between model providers with minimal friction, a pricing war makes it cheaper to run more agents, more often, with longer context windows.
The risk sits on the other side of the ledger. If frontier model companies compress margins to win market share, the question becomes whether they can sustain the compute infrastructure that agents depend on. TradingKey’s analysis compared the dynamic to the electric vehicle price war: short-term market share gains, but long-term margin compression that challenges commercial viability. Some analysts cited in that report argue significant B2B cuts are unlikely given compute, storage, and networking costs, with more aggressive reductions concentrated in consumer-facing products.
The IPO Pricing Paradox
Both companies face an unusual bind. Cutting token prices boosts user acquisition metrics that look good in an S-1 filing. But it also signals to public market investors that pricing power is eroding faster than expected. OpenAI’s $852 billion valuation and Anthropic’s $965 billion valuation both assume sustained revenue growth from API and subscription products. A race to the bottom on token pricing compresses the margins those valuations depend on.
For agent builders, the near-term calculus is straightforward: lower token prices reduce deployment costs and make previously uneconomical agent workflows viable. The longer-term question is whether the companies providing those tokens can survive the margins they are creating.