OpenAI is weighing steep cuts to the prices it charges for tokens, the units AI firms use to bill for their products, the Wall Street Journal reported Wednesday, citing people familiar with the matter. The talks are still in flux. OpenAI is reportedly preparing for similar cuts it expects from Anthropic. The aim would be to pull users toward ChatGPT as competition between the two labs intensifies.
OpenAI currently sells consumer access to its GPT-5.5 models in tiers of $8, $20 and $100 or more per month. Anthropic charges $17 a month for Claude Pro on an annual plan and $100 or more for Claude Max. CEO Sam Altman has signaled the shift. He said at a recent event that cost has become a major concern for business customers and that the company would look for ways to deliver more value for less. The reported plan centers on token pricing rather than headline subscription tiers.
The report lands during a sprint toward public markets. OpenAI confidentially filed for an IPO on June 8, days after a similar filing from Anthropic. Altman has told staff he expects OpenAI to go public within the next year, a later timeline than the October debut some had anticipated. The two are closely matched on paper. Anthropic closed its Series H round on May 28 at a $965 billion valuation, narrowly ahead of OpenAI’s $852 billion mark from March. ChatGPT became the first app to reach 1 billion monthly users in May, about three years after its November 2022 launch, faster than Google Maps managed, according to Sensor Tower.
Why It Matters
- A cut by either lab could force a quick response, since enterprise customers can switch providers easily and the products are largely interchangeable.
- Discounting would squeeze margins just as both companies court public investors.
- Analysts warn that competing mainly on price suggests model quality alone is no longer a strong enough differentiator.
- Cheaper tokens could speed enterprise adoption and strengthen OpenAI’s growth story for its listing.
How We Got Here
Both labs are absorbing heavy losses driven by computing costs, with combined 2026 spending on compute, training and operations estimated near $65 billion. Anthropic aims to reach breakeven around 2028, while OpenAI’s path runs roughly two years longer.
By one WSJ estimate, OpenAI plans $121 billion in compute spending in 2028 and could post an $85 billion loss that year alone. The price talks also follow Anthropic’s release of its Mythos-class models. Its first general-release Mythos model, Claude Fable 5, costs about double its Opus models, making it the company’s priciest tier and underscoring how the rivals are splitting between premium and value strategies.
