OpenRouter developers shift to cheaper AI models as prices fall

AI token costs have dropped from $5.63 to $0.21. Developers on OpenRouter have noticed, and they are voting with their API keys.

The platform’s traffic now leans heavily toward cheaper, “good enough” models rather than the flashiest frontier releases.

The numbers behind the downshift

Venture capitalist Tomasz Tunguz has analyzed OpenRouter’s usage. He found that 84% of the tokens processed on the platform run through non-frontier models.

According to Tunguz’s analysis, they deliver about 77% of the performance of leading models at just 2.5% of the price.

The price gap is stark. Non-frontier models on OpenRouter average roughly $0.50 per million tokens. Claude Fable 5, a top-tier model, costs $20 per million tokens as of August 2026.

That is a 40-to-1 spread.

At the bottom of the market, prices have fallen even further. Some models now cost between $0.03 and $0.21 per million output tokens.

Price cuts are driving usage, not killing revenue

OpenAI cut prices on GPT-5.6 Luna by approximately 80%.

The result was a 14-fold jump in usage for that model. Revenue from it rose 34%.

Chinese models have accounted for over 30% of US company token consumption on OpenRouter, peaking at 46%.

In early 2025, that share was about 4.5%.

How OpenRouter fits into the picture

OpenRouter acts as a middle layer between developers and AI providers. It aggregates more than 400 models from over 80 providers behind a single interface.

The platform processes hundreds of trillions of tokens every month.

One feature, called :floor routing, automatically finds the cheapest provider for a given model.

OpenRouter says it charges no markup on provider rates. Instead, it applies a 5-5.5% fee when users purchase credits.

Batch API offerings add another lever. They promise additional savings for developers willing to process requests in bulk rather than in real time.