# Artificial Analysis Index is non-representative of real-world performance. Tested Kimi 3.1 Pro through OpenRouter, the number 4 model in AAI, it's not even close to MiMo 2.6 or DeepSeek 4.1 Flash — retained record

> Retained reddit record from reddit:LocalLLaMA.

Canonical: https://highsignal.app/data/records/21f4252804f4a9c8

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:LocalLLaMA)  published 2026-10-25 03:04 ingested 2026-10-26 02:40

[original source ↗](https://www.reddit.com/r/LocalLLaMA/comments/1xercio/artificial_analysis_index_is_nonrepresentative_of/)

## retained text



```
It seems Artificial Analysis Index is very easy to game, Kimi 3.1 Pro #4 on AAI is super bad in real world use cases, not even close to DeepSeek 4.1 Flash.
```



document fetched 2026-10-26 00:38 · canonical https://reddit.com/r/LocalLLaMA/comments/1xercio/artificial_analysis_index_is_nonrepresentative_of
