A new project at jock.pl tracks the "staleness" of 20 current AI models by monitoring two critical dates: the release date and the training cutoff. While the release date marks when a model is made available to the public, the training cutoff is the date when the model's training process stopped.
New Tool Tracks AI Model Staleness by Comparing Release Dates and Training Cutoffs
The gap between these two dates indicates how much information the model lacks at the moment of its launch. For example, the tool notes that GPT-6 Astra was released on September 3, 2026, but its training data stops at April 30, 2026, meaning it was already five months behind upon release.
The data shows that out of 20 models across 8 labs, only 9 have a training cutoff that the lab actually publishes. Recognized labs that have published cutoffs for at least one model include Anthropic, Google DeepMind, Meta, and OpenAI. The tool serves as a reference for users to verify model knowledge, noting that models often lack updated information despite having web search capabilities.
Sources
- Show HN: How Stale Is Your AI? Release age and training cutoff for 20 models （Hacker News Frontpage, 2026-09-16）
- wiz.jock.pl