As of 2026-08-13, by what percentage does adopting each successive frontier model release raise research productivity at the largest AI lab by revenue, per unit of Epoch Capabilities Index improvement?

Final estimates: p10=0.9, p25=1.8, p50=3.2, p75=5.8, p90=9.8. Initial logic and parameters are validated regarding the log-linear percentage increase in overall research-output value per ECI point elasticity.institute. Standard processing accounts for the wedge between automated coding throughput and true R&D progress 44 sources, establishing overall organizational progress parameters 55 sources. Aligning the translation of capability gains into research productivity with structural friction models and tying this elasticity to consistent projections of near-term absolute AI R&D value uplift tightened our central expectation slightly, aligning with baseline projections of around 16 points of ECI progress in 2029 and sub-exponential compute growth . The lower bounds (p10 of 0.9%, p25 of 1.8%) reflect the likelihood that self-reported value gains face steep diminishing returns. Conversely, the right tail remains thick (p75 of 5.8%, p90 of 9.8%) due to unobserved internal usage 2 sources and the potential for convex returns once models cross key reliability thresholds, which would trigger rapid recursive acceleration .