OpenAI's GPT-4 maintained the top position on the Epoch Capabilities Index for approximately 12 months, according to Epoch AI researcher Jaeho Lee. This period of dominance is significantly longer than any model has managed since GPT-4's release. The ECI is a composite measure of language model performance, tracking how long models remain at the top.
Since Claude 3 Opus overtook GPT-4 in February 2024, the leadership on the ECI has changed hands 17 times. The median time a model stays at the top is about seven weeks, according to the data. This indicates that the competition among top models has intensified, with shorter durations of dominance and smaller capability jumps between transitions compared to GPT-4's era.
The chart highlights two key trends: the prolonged lead of GPT-4 and the accelerated pace of innovation in recent months. While GPT-4 was a genuine outlier at launch, today's models face a more competitive landscape where no single model can maintain a comparable lead for an extended period. Source: thedecoder
