Move the sliders — the ranking updates instantly. All weights start at 5.
As of July 2026. Capability: benchmark scores (HLE, GPQA Diamond, SWE-bench). Price: output tokens, logarithmically normalized (DeepSeek V4 Pro = 100, Fable 5 = 0). Openness: open-weight availability. Speed and German proficiency of Chinese models: estimates due to lack of independent tests. No warranty.