AI Radar

Noteworthy AI conversations and writing, summarized from publisher notes.

Are Open Models Catching Up?

· SemiAnalysis

The article examines the progression of open versus closed-source AI models, categorizing their development into three distinct eras: early scaling (2022-2024), reasoning (2024-2025), and agentic (2025-Today). It asserts that the gap between open and closed models is consistently narrowing, with open-source models taking approximately half as long to catch up to frontier closed-source models in each subsequent era. The analysis uses curated benchmarks for each era, acknowledging their limitations and the impact of productization on user experience. Despite the rapid closing of the gap, the article suggests that this trend is not necessarily bearish for frontier models, hinting at further analysis in the full text.

AI Radar wrote this summary from the publisher’s notes with AI, not from a transcript. How it works.