Begun the Token Pricing Wars Have
Summary
The foundation model market is experiencing intense token pricing competition, driving down profitability despite surging demand. Economists have long predicted this "Bertrand supertrap" due to the limited differentiation between models for most use cases. Current subscription models, like a \$200-a-month plan offering many thousands of dollars in tokens, are unsustainable, leading providers to lose money. Users exploit generous limits by creating multiple accounts rather than paying marginal token prices. Anthropic's attempts to shift its token-intensive Fable 5 product off subscriptions faced setbacks when OpenAI released GPT 5.6 Sol without similar restrictions, forcing Anthropic to extend deadlines. Providers are hoping Recursive Self-Improvement (RSI) will create switching costs, locking in users and mitigating competition, but this remains uncertain. The industry faces a dilemma: it needs users to economize tokens, but current pricing encourages overuse, hindering user acquisition if restrictions are imposed.
Key takeaway
For AI Product Managers evaluating pricing strategies, the current token subscription models are unsustainable due to intense competition and user behavior. You should anticipate continued price pressure and user exploitation of generous plans. Consider exploring models that either drastically reduce inference costs or effectively integrate provider-side token management and pricing signals to encourage economization, rather than relying on user-driven efficiency or uncertain future lock-in from Recursive Self-Improvement.
Key insights
Foundation model providers face an unsustainable token pricing war driven by competition and user behavior.
Principles
- Price competition drives similar products to marginal cost.
- Generous subsidies lead to user exploitation.
- Recursive Self-Improvement (RSI) aims to create user lock-in.
In practice
- Users exploit generous token limits by creating multiple accounts.
- Providers attempt to create switching costs via Recursive Self-Improvement.
- Economizing tokens is critical but difficult to enforce with current models.
Topics
- Token Pricing
- Foundation Models
- Price Competition
- Subscription Models
- Recursive Self-Improvement
- Inference Costs
Best for: Investor, Entrepreneur, CTO, Director of AI/ML, AI Product Manager, Executive
Related on AIssential
See Counsel's argued verdicts on the open AI decisions leaders are weighing →
Editorial summary, takeaway, and curation by AIssential. Original article published by Joshua Gans' Newsletter.