Skip to main content
MANIFOLD
Will the OpenRouter LLM input:output token ratio exceed 100 in June 2026?
10
Ṁ100Ṁ249
Jun 7
25%
chance
4

This will resolve by taking the most popular model by $ volume (i.e. token volume * input token price) on OpenRouter, and considering a 7 day unweighted average of “prompt tokens divided by (output tokens + reasoning tokens)”

For example, currently, the top model by the above criteria is Sonnet 4, https://openrouter.ai/anthropic/claude-sonnet-4/activity, and the average would be about 50.

This will resolve using data from Jun 1-7 2026.

Market context
Get
Ṁ1,000
to start trading!
Sort by:
sold Ṁ9 NO

I don't understand how you are calculating the ratio. Can you explain it with more detail?

@Eternal On this page: https://openrouter.ai/anthropic/claude-opus-4.8/activity
Add together "Completion" and "Reasoning" tokens.
Then divide "Input" tokens by that number.

Currently, it looks under 100, something like 50-80

Ultimately this question seeks to ask: "There's currently a trend where LLMs are reading more content per unit of output, will this trend continue?"