# @MikePFrank — 2026-08-19

♥10 ↻2 · https://x.com/MikePFrank/status/2090204245635035173

Some data for models scoring at or above Qwen3.8 27B on the AAI Index, sorted by "cost efficiency" defined as AAI's index score divided by AAI's cost-per-task. Speed (tps throughput) is the best value sampled from OR at the moment I happened to look.

Interestingly, Luna is the still most cost-efficient model (if AAI can be believed), and DeepSeek V4 Flash is 2nd.

Qwen-3.8 27B is only 4th on this metric, but of all these models, it's the only one you could run on a laptop.

![photo](../../../media/HQHkdurWAAAMDU9.png)
tags: author:mikepfrank, has-image, kind:image, kind:tweet, on:qwen-3-8-27b, year:2026
cited on: _dossiers/qwen-3-8-27b.md, qwen-3-8-27b