What it means for leaders

Capability that was the best available a year ago now sells for a fraction of the frontier price. Revisit the model behind every production workload at least yearly: the same quality is usually much cheaper.

What would change the reading: How many models qualify, and whether open-weight models set the price.

The breakdown

October 25, 2026.
  • DeepSeek V4.1 Flash$0.27 per million tokens
  • GLM-5.3-Flash$0.24 per million tokens
  • Claude Haiku 5.5$0.20 per million tokens
  • GPT-6 Luna$0.20 per million tokens
  • DeepSeek V4 Flash 0731$0.07 per million tokens

Every reading

DateValueChangeDetailSource
$0.07 per million tokens—DeepSeek V4 Flash 0731 (matches GPT-5 Pro)Source

Definition and method

What it measures
The lowest list price, per million tokens blended three input to one output, of any model whose ECI is at least that of the best model released up to twelve months earlier.
Population
Models scored on the Epoch Capabilities Index and listed on OpenRouter
Updated
Each update
Source
OpenRouter, model list prices · Provider list prices, attributed; Epoch AI, Epoch Capabilities Index · CC BY 4.0

Caveats

  • List prices exclude volume discounts and batch pricing.
  • Only models matched by name between Epoch AI and OpenRouter are priced.

Cite this

Enterprise AI in Numbers, "What last year's most capable AI costs today", https://lucianoayres.com/ai-brief/data/price-of-last-years-frontier/, retrieved October 25, 2026. Data: OpenRouter, model list prices and Epoch AI, Epoch Capabilities Index.

Download every indicator (CSV) · Methodology