HOW FAR BEHIND ARE OPEN MODELS?

Summary from the TLDR AI newsletter

Open models are generally not as capable as the best closed models. However, they aren't too far behind, with tests showing that they are only four to six months behind on public benchmarks. The gap was the smallest around the time of DeepSeek R1. It has since been growing.