Benchmarks & Speed from Artificial Analysis
Intelligence Index, Coding Index, Math Index, GPQA, HLE, IFBench, TAU2, LiveCodeBench, TTFT, tok/s, and more.
Pricing & Features from OpenRouter
Input/output/blended cost, context length, tool support, vision support, cache pricing.
VRAM Estimates calculated from parameter count.
Q4: active_params × 0.55 + 2 GB overhead.
FP16: active_params × 2 + 4 GB overhead.
MoE models use active parameters, not total.
Scores are weighted composites (0-100) computed client-side. Adjust via Weights panel and presets. "Score+Code" includes coding; "Score Gen." excludes it.
Table
Frontier
Chart
Compare 0
Pareto Frontier Models
Top performers for their cost
No models selected
Check models in the table to compare them side by side.