Plots
Interactive charts generated from my AI-forecasting data pipeline. Dates are when the underlying data was last refreshed.
- BECI capability timeline Every model's BECI capability score against its release date, colored by organization, with bootstrap CIs — the capability index behind the other plots.
- BECI vs inference cost Capability (BECI) against measured inference cost per task, stitched across ~130 leaderboards with an IRT-style factor model. Release-date slider recomputes the Pareto frontier.
- BECI vs task latency Capability against measured wall-clock time per task (deliberation + agent steps, not serving speed), same factor-stitch as the cost index.
- BECI vs tokens spent Capability against consumed tokens per task (total where reported, else input+output), anchored to tokens per AIME 2025 problem.
- BECI benchmark browser Every benchmark's raw scores over time with the BECI logistic fit overlaid — searchable, with per-benchmark fit diagnostics.
- Model size estimates Parameter-count estimates for frontier models from the size-signature analysis.