AI Model Prices Are Collapsing: What It Means for Enterprise

Key Takeaways

• Frontier models command only 3.7% of spending despite being the most capable and expensive options available
• Open-source models now run 86% cheaper than closed models while capturing the majority of token volume on major gateways
• The real competition is in the messy middle — multi-step workflows where price and capability balance matters most
• Demand follows a normal distribution, not a pyramid: the fat middle prioritizes intelligence per dollar, not cutting-edge capability
• Price cuts are accelerating: Anthropic held prices steady for five releases until yesterday, while OpenAI has cut Luna by 80% then 50% again

The Price War at the Frontier

Yesterday, Anthropic released a new model and cut its price. Ninety minutes later, OpenAI did the same. This race to the bottom is reshaping AI economics.

In June, Anthropic set the frontier price at $10 and $50 per million tokens with Fable 5. OpenAI answered with GPT-5.6 Sol at $5 and $30 — matching capability at a third of the cost. The Anthropic Opus line, which had remained stable across five releases at $5 and $25 per million tokens, finally broke yesterday with its first price cut.

At the low end, the collapse is even more dramatic. OpenAI cut Luna by 80% in July, then slashed it another 50% yesterday.

Open Models Are Winning on Price

Closed-source rivalry isn't the only force driving prices down. Open-source models now run a majority of token volume on gateways that publish data — at an 86% discount to the blended price of closed models.

Large customers are pushing further by fine-tuning. Cursor's Composer 2 fine-tuned Kimi K2.5, an open-weight base, cutting overall cost 86% against its previous in-house model. Harvey achieved similar results, cutting cost per cell 55% against Sonnet 5 while scoring higher than Fable 5.

The Frontier Models Aren't Where the Money Is

Despite being the most capable, frontier models command surprisingly small market share. Anthropic's Fable 5.1 captured only 3.7% of gateway spending in its first twelve days. Its predecessor peaked at 13.2% when access was restored in July, then fell to 4.9% when Opus 5 shipped at half the price.

Among large corporate accounts, frontier model consumption dropped from 53% of tokens in early August to 45% by September.

The distribution of demand isn't a pyramid with a wealthy peak. It's a normal distribution with a fat middle. Enterprises buy intelligence per dollar, not maximum capability.

Why the Middle Tier Matters Most

Most business AI use is the messy middle: multi-step workflows that need a smart enough model at a price a company can afford. This is the most important part of the market today and where competition is fiercest. The mid-tier claims 40% of spending and 30% of tokens.

As intelligence per dollar explodes, enterprises upgrade to cheaper tiers that now meet their fixed requirements. The question that will determine AI economics is whether token volume eventually shifts entirely to commodity models.

Conclusion

AI pricing has collapsed faster than almost any transformative technology in history. The real competition isn't at the frontier — it's in the middle tier where enterprises make purchase decisions. Whether the market consolidates around commodity models will shape the entire AI economy going forward.

---

Original source: The Most Important Market in AI is the Middle

powered by osmu.app