MikePFrank Ranks Model Cost-Efficiency via AAI Data: Luna Leads, Qwen 3.8 27B Fourth

From August 20 to 21, MikePFrank posted a series of analyses ranking models by cost-efficiency—Intelligence Index score divided by per-task cost—based on public Artificial Analysis Index data, correcting his calculation method mid-analysis. The conclusion: Luna is the most cost-efficient model, DeepSeek V4 Flash follows, and the newly released Qwen-3.8 27B ranks fourth, though it is the only model on the list that runs locally on a laptop and sits on the Agent cost-efficiency Pareto frontier, where Claude does not.

Confirmed

Among models scoring at or above Qwen-3.8 27B, Luna has the highest intelligence-per-unit-cost, with DeepSeek V4 Flash second and Qwen-3.8 27B fourth.
Qwen-3.8 27B is the only model on the list runnable locally on a laptop.
The author corrected the "unit-cost agent index" column, recalculating it based on per-test cost; the updated table changed only a few rankings.
In the Pareto frontier chart of agent cost-efficiency built from AA data, Qwen-3.8 27B lies on the frontier while Claude does not.
AAI provides its own Pareto frontier charts of Intelligence Index score versus cost.

Why it matters

The author argues the Pareto frontier view shows performance-versus-cost tradeoffs more intuitively than rankings, aiding model selection across budgets.
The analysis folds local deployability into the value discussion, highlighting the significance of small models like Qwen-3.8 27B for local developers.

2026-08-20 ~ 2026-08-21 · 6 related posts

Primary sources

Luna leads cost-efficiency, Qwen 27B stands out as laptop-viable option — MikePFrank · 2026-08-20 06:28 (UTC+08:00)
Correcting Agentic Index: Introducing Cost-Per-Test and Pareto Frontiers Analysis — MikePFrank · 2026-08-20 08:26 (UTC+08:00)
Efficiency Chart: Qwen-3.8-27B Hits Pareto Frontier, Outperforms Claude — MikePFrank · 2026-08-21 00:37 (UTC+08:00)

[source] Luna leads cost-efficiency, Qwen 27B stands out as laptop-viable option — MikePFrank · 2026-08-20
Luna Remains Most Cost-Efficient, Qwen 3.8 27B Ranks Fourth — MikePFrank · 2026-08-20
AAI Publishes Pareto Frontier for Intelligence Index vs Cost — MikePFrank · 2026-08-20
[source] Correcting Agentic Index: Introducing Cost-Per-Test and Pareto Frontiers Analysis — MikePFrank · 2026-08-20
Agent cost-efficiency plot: Qwen hits Pareto frontier — MikePFrank · 2026-08-20
[source] Efficiency Chart: Qwen-3.8-27B Hits Pareto Frontier, Outperforms Claude — MikePFrank · 2026-08-21