Story · vc·
Mento's Startup Agent Benchmark: 14 AI models research startups well, but none beat a headcount-growth rule at ranking them
Mento, a venture fund, published its Startup Agent Benchmark (SAB), which as of October 2026 compares 14 models from four vendors on 960 startup research tasks and 101 ranking cohorts, half on synthetic companies and half on a frozen snapshot of Mento's own company data. Claude Opus 5.5 led the research tasks with 97.2 of 100, ahead of Claude Fable 5.1 at 96.8 with overlapping intervals, at less than a third of Fable 5.1's cost per task. Every model gave a value for 9% to 33% of facts Mento had deliberately removed from the corpus. On ranking, all 13 models that completed the cohorts scored below a simple rule that orders companies by past headcount growth.
A VC's own test suggests agents are ready for diligence research with source checks, but not for picking which startups will grow.
Source
Read at Mento.vc ↗Habr post by the Mento team (Russian, Oct 29) ↗
Mento.vc · mento.vc · Found in the news · source checked Nov 1, 2026
Related tools
Spot an error? Submit a correction
One email a week. Free.