LLM Efficient Frontier
tokens/s vs quality
Compare
Scale: Lin
All Categories
Reasoning (R1, QwQ)
Coding (Qwen-Coder)
Dense General
MoE
Edge (≤ 3.8B)
All Families
All Sizes
≤ 9B
10B - 35B
70B+
All Release Years
Tensor parallel
+6 mo projection
Show models >2 yrs old
Hide OOM
Allow Offload
Pareto Frontier
Frontier Models
All Models
Click column header to sort • Click row to inspect
Methodology: Memory Bandwidth & Inference Performance Model
▾