Artificial Analysis Progress Index v1
Indexes
Artificial Analysis Progress Index v1 places language models on one capability scale with an item response theory fit over every Artificial Analysis evaluation a model has, linking models through shared tasks so models tested on different evaluation sets remain comparable over time. Each effort variant is fit separately; the scale is anchored at o3-mini (high) = 220 and GPT-5.5 (xhigh) = 320 and has no fixed range. Artificial Analysis refits it as models are added. Kept outside the capability fit as a diagnostic, but built from Artificial Analysis evaluations that the fit also reads. Variants with executed model fallback on any evaluation are excluded from single-model observations.