Luna Remains Most Cost-Efficient, Qwen 3.8 27B Ranks Fourth
MikePFrank · x · 2026-08-20
Analysis based on the AAI Index reveals that among models scoring at or above Qwen 3.8 27B, Luna remains the most cost-efficient model per unit cost, followed by DeepSeek V4 Flash. Qwen 3.8 27B ranks fourth in cost efficiency. The author notes that score differences among top models are minimal, causing the cost-efficiency ranking to align closely with the lowest-to-highest cost-per-task ranking.
More from Models
- Grok Bot Ran a 12-Hour Data Analysis Nonstop, Handling API Rate Limits Itself — Daniel_Farinax · 2026-10-16
- One year of Recursive Language Models: GPT-5-mini beats GPT-5 by 110% on 132k-token tasks, cheaper too — a1zhang · 2026-10-16
- Users Report Claude Models Suddenly More Compliant After Forced Compaction — RealSharpNinja · 2026-10-16
- Prediction: Opus 5.5-level models on laptops by 2027, backed by Qwen 27B data — cephaloform · 2026-10-16
- Users Say Krea 2 Output Degraded Sharply Since Krea 3 Beta Launched — szywis · 2026-10-16
- LLM-as-a-Verifier lifts Nemotron 3.5 Super by 9-10 points on agentic benchmarks — simonguozirui · 2026-10-16