The study presents the distribution of frequency for benchmarks reported in 2026 by the most advanced laboratories, highlighting that HLE, SWE-Bench Pro, DeepSWE and Terminal-Bench achieved universal adoption, DeepSWE was reported by all 12 laboratories, and OpenAI created the majority of the popular benchmarks; this trend affects the influence of benchmark makers and competition.
AI-generated text
Frequency analysis of benchmark usage by frontier laboratories in 2026
The study presents the distribution of frequency for benchmarks reported in 2026 by the most advanced laboratories, highlighting that HLE, SWE-Bench Pro, DeepSWE and Terminal-Bench achieved universal…



