> TLDR: before reaching for the cheapest-looking model, check its cost per task, not its price per token.

from AA:

> Pricing for DeepSeek V4.1 Flash (Max) is $0.30 per 1M input tokens (moderately priced, median: $0.43) and $1.20 per 1M output tokens (moderately priced, median: $1.68). On average, it costs $0.27 per task to evaluate DeepSeek V4.1 Flash (Max) on the Intelligence Index.

I feel like this completely overlooks major details like the fact that open router offers prices way below this?

DeepInfra for instance right now, 8bit

in: $0.14

out: $0.4

cache: $0.0042

is this not a major reason to prefer more open models?