OpenAI chairman says token pricing will vanish in 12 months

OpenAI Chairman Bret Taylor predicts the AI industry's token-based pricing model will be replaced by outcome-based pricing within 12 months.

OpenAI Chairman Bret Taylor said the AI industry's token-based pricing model will be replaced by a pay-per-outcome system within 12 months, as companies push for clearer returns on their AI investments.

"I believe where the world is going is paying for outcomes," Taylor, who also co-founded AI startup Sierra, said in a CNBC interview Monday.

Token costs have prompted companies to rein in AI spending this year, with CFOs scrutinizing whether they are getting a return on investment. Taylor compared the current market to the early internet, when building a website cost substantially more. He pointed to Ramp's token spend tracking tool and legal AI startup Harvey as examples of companies managing token complexity for clients, rather than forcing enterprises to track tokens themselves.

The shift could reshape how enterprises budget for AI, moving from per-token accounting to subscription or outcome-based pricing. Taylor predicted IT departments will become "really sophisticated" about AI applications within 12 months, with different tools for marketing, engineering and other functions. "You just don't need to think about the word token at all," he said.

Kimi K3 and the Cost Question

Taylor struck a cautious tone on Moonshot's Kimi K3, a model released last week that has drawn Silicon Valley attention for rivaling OpenAI and Anthropic's capabilities at a lower reported cost. "Is it cheaper to use is the most important part for anyone considering it," Taylor said. "And I think the jury's out on that."

The skepticism reflects a broader industry debate over whether open-weight models can sustain their cost advantage at scale. Frontier models from OpenAI and Anthropic remain more token-efficient, Taylor argued, suggesting that headline pricing comparisons may not capture real-world inference costs. Kimi K3's reported performance on key benchmarks has not been independently verified against the test conditions used by OpenAI's GPT-4o or Anthropic's Claude 3.5 Sonnet, leaving enterprise buyers without a clear basis for comparison.

Who Wins, Who Loses in the Pricing Shift

The transition from token-based to outcome-based pricing carries implications across the AI value chain. Application-layer companies — Sierra, Harvey, Ramp and others that build on top of foundation models — stand to benefit as enterprises shift from managing token budgets to paying for specific business outcomes. These companies can absorb token cost fluctuations and present a simplified pricing structure to customers.

For model providers including OpenAI, Anthropic and Google's DeepMind, the move could compress margins if outcome-based pricing proves less profitable than per-token billing. OpenAI's API pricing currently ranges from $0.01 to $0.03 per 1,000 tokens for its GPT-4 class models, depending on input versus output. A shift to outcome pricing would transfer cost risk from the enterprise to the model provider, potentially squeezing margins at the infrastructure layer.

Moonshot's Kimi K3, if its cost advantages are confirmed, could accelerate pricing pressure on frontier model providers. The Chinese startup has not disclosed the full methodology behind its performance claims, leaving investors and enterprise buyers without a clear basis for comparison. Taylor's "jury's out" assessment suggests the market should treat cost claims with caution until independent benchmarks are available.

Investment Implications

For investors, the pricing model transition creates divergent incentives. Companies that build AI applications for specific verticals — legal, finance, marketing — may see improved adoption as enterprise buyers face simpler pricing. Conversely, pure-play model providers could face margin compression if outcome-based pricing becomes the industry standard.

The broader AI infrastructure ecosystem — GPU providers like Nvidia, cloud platforms including Amazon Web Services and Microsoft Azure, and data center operators — is less directly affected by the pricing model shift, since their revenue depends on compute consumption rather than token pricing. However, if outcome-based pricing drives higher AI adoption, compute demand could accelerate, benefiting infrastructure providers regardless of the pricing mechanism at the application layer.

Taylor's prediction, if realized, would mark a fundamental shift in how AI is bought and sold — one that could make the word "token" irrelevant to enterprise buyers within a year.

This article is for informational purposes only and does not constitute investment advice.