Gap Between Open Source AI and Closed Source Cutting-Edge Models Reduced from 12 Months to 4 Months
According to crypto and tech analyst Miles Deutscher citing Epoch AI data, the capability gap between open source AI models and closed source cutting-edge large models has shrunk from about 12 months two years ago to approximately 4 months now.
Open source has caught up from Llama 2-70B to GLM 5.2 and Kimi K2.6, with expectations that in another 6 months, top-tier intelligence at the Claude Fable level will be available for free download in open source form.
This trend weakens the moats of closed source giants and may reshape the AI business landscape and computing power competition.
Source: Public Information
ABAB AI Insight
Miles Deutscher has been tracking AI development for a long time, and this assessment of the reduced gap is based on the Epoch AI index, similar to the rapid iteration trajectory of the Llama series.
In terms of capital pathways, the rise of open source drives resources towards open source model training and distribution platforms, motivated by lowering entry barriers and expanding the ecosystem.
Similar to the past impact of open source software on commercial software, the current AI landscape is experiencing a rapid decline in closed source advantages and the early stages of open source accessibility.
Essentially, this is a technological substitution: the acceleration of open source is democratizing capabilities, concentrating capital towards open source ecosystems and low-cost computing solutions, reshaping AI business models and weakening the pricing power of closed source giants.
ABAB News · Cognitive Law
The reduction in the gap signifies the erosion of moats: from December to April, the rise of open source is unstoppable.
In 6 months, top-tier open source will be free: Claude-level intelligence, available for public download and use.
The open source ecosystem is reshaping the AI landscape: the retreat of closed source advantages and a shift in business competition towards ecosystem building.