The AI model market is now pricing in two directions at once. Anthropic released Fable 5 on July 24 at $50 per million output tokens, double the price of Opus 5. Five days later, OpenAI cut GPT-5.6 Luna’s price by 80 percent, permanently.
VC analyst Tomer Tunguz frames the tension with a question, can the industry keep sustaining Jevons’ paradox, the pattern where cheaper AI drives so much more usage that total compute demand keeps growing? The pricing moves make that question concrete.
What the Numbers Actually Show
Fable 5 at $50 per million output tokens is a ceiling-raising move. It prices a premium reasoning tier well above the previous flagship, and it signals that Anthropic believes there is a segment willing to pay a large premium for frontier capability.
Luna’s 80 percent cut moves in the opposite direction. It is a floor-lowering move aimed at volume, developer adoption, and price-sensitive workloads.
The two moves are not contradictory, they are complementary tiers in a maturing market. What they show is that model pricing is splitting: a high-margin frontier tier at the top, and a commoditized volume tier below it.
The Tension Beneath the Headlines
Tunguz’s question about Jevons’ paradox is the deeper issue. If lower prices grow usage fast enough, total compute spending keeps rising even as unit prices fall. That is the bull case for AI infrastructure, and it is why GPU supply remains the constraint.
But there is a countervailing force. The paradox only holds if demand is elastic, and it depends on capacity actually being available to serve the new usage. If demand grows faster than compute supply, the result is not more volume, it is rationing and price pressure at the top, which is exactly the tier Fable 5 occupies.
The evidence on both sides is limited. Executive statements that supply cannot meet demand are common, but they are incentives-laden claims from companies selling compute. The pricing moves themselves are the more reliable signal, and they say the market is hedging both ways.
What This Means for Buyers
For developers and businesses, the practical implication is a widening choice. The volume tier is getting cheaper, which lowers the cost of experimentation and high-volume work. The frontier tier is getting more expensive, which raises the bar for when the premium is worth paying.
The strategic takeaway is to price your workloads, not your models. Know which of your tasks genuinely need frontier output, and route the rest to the volume tier. The market is now structured to reward exactly that discipline.
What Remains Unresolved
The open question is whether the price split is stable. If Fable 5’s premium holds, it validates ceiling-raising. If competition forces it down within a quarter, it suggests the premium tier was overpriced from the start.
The second unresolved question is the capacity one. The paradox argument depends on supply growing to meet demand. If it does not, the ceiling keeps rising, and the floor stops falling. That combination would be a real departure from the last two years of AI economics.
Related Reads
- [Claude API Pricing 2026](https://getaibest.com/claude-api-pricing-2026-complete/)
- [Claude Opus vs Sonnet Pricing](https://getaibest.com/claude-opus-vs-sonnet-pricing/)
- [GPT-Live: Real-Time Audio Architecture](https://getaibest.com/openai-gpt-live-real-time-audio-architecture/)