Hacker News
SemiAnalysis Measured the Open-Weights Catch-Up Time and Found It Halves Every Era, Now Down to Under Five Months
SemiAnalysis published on August 21 that the open vs closed gap moves in cycles and closes roughly twice as fast each generation. Era 1 scaling: GPT-3.5 Turbo 75.7 vs Llama-2-70B 39.9, closed when DeepSeek V3 matched GPT-4o in December 2024. Era 2 reasoning: o1-preview led DeepSeek R1 by 12.1 points, closed by the R1-0528 checkpoint in an 8.5-month window. Era 3 agentic: Kimi K2.6 passed Opus 4.5 in 4.8 months and GLM-5.2 cleared GPT-5.2 in 6. The authors flag their own caveat that benchmarks miss productization.
Source
↳ Follow the thread