Agent Arena Data Reveals 93:1 Input-to-Output Token Ratio

1 reports archived1 independent sourcesupdated 11/3/2026, 03:08:00

Synthesis & Latest Updates

1 Sources Cross-Validated

Analysis by Arena.ai reveals that top agents process ~93 input tokens for every generated token. This high context consumption explains the industry race to lower cached input costs amid an emerging 'hidden AI price war'.

LATEST/Analysis by Arena.ai reveals that top agents process ~93 input tokens for every generated token. This high context consumption explains the industry race to lower cached input costs amid an emerging 'hidden AI price war'.

Hourly heat curve

2 fully observed hours

Fewer than 3 fully observed hours; trend not plotted yet.

Coverage timeline

Total 1 reports · Latest first

Arena.ai (@arena)T1Nov 3, 03:08·68 pts
Agent Arena Data Reveals 93:1 Input-to-Output Token Ratio
- Top agents exhibit a ~93:1 input-to-output token ratio, indicating heavy reliance on large context windows for reasoning.
- High input volume drives model providers (e.g., Anthropic, Google) to accelerate optimizations in KV Cache and input caching to reduce marginal costs.
View Related Reports & Analysis Original Source