GPT-4 did something that now looks increasingly rare: it stayed ahead. On the Epoch Capabilities Index (ECI), OpenAI's GPT-4 held the top position for about a year, while newer leading models have been replaced far more quickly.
The shift is not just about one model losing ground. It points to a faster, tighter race among AI labs, where the best-performing language model can become yesterday's leader within weeks.
GPT-4's lead became the exception
The ECI is a composite measure of language model performance. According to Epoch AI researcher Jaeho Lee, GPT-4's time at the top lasted far longer than any model that followed it.
OpenAI's o1 had the second-longest stay in first place, but that lead lasted just over three months. That was less than a third of GPT-4's run.
That comparison matters because it shows how unusual GPT-4 was at launch. A model that can remain unmatched for about a year is not merely taking first place; it is creating a gap that competitors need significant time to close.
The top position now changes much faster
The picture changed after Claude 3 Opus dethroned GPT-4 in February 2024. Since then, the lead on the ECI has changed hands 17 times.
The median stay at number one per model has been about seven weeks. In practical terms, that means the top of the AI model rankings has become much less stable.
For anyone watching language model performance, this creates a different kind of market signal. A new leader may still be important, but its lead is less likely to define an era in the way GPT-4's did.
What the ECI trend suggests about competition
The same chart can support two connected readings. First, it shows how long rival labs needed to match GPT-4. Second, it shows that competition has intensified since that period.
No later model has held a comparable lead. Instead, the field has moved into a pattern of quicker transitions, where one model overtakes another and is then overtaken in turn.
The article also notes that the capability jumps between transitions are faster but smaller compared with GPT-4 and the period that began with reasoning models like o1-preview in fall 2024.
That distinction is important. Faster movement does not automatically mean every new leader creates a dramatic break from what came before. It can also mean that competitors are clustered more closely together.
Why shorter reigns matter
For developers, companies, and researchers, a shorter stay at the top changes how AI model leadership should be interpreted. A model may lead on the ECI, but the median pattern suggests that leadership may be temporary.
This does not make the ranking meaningless. It makes timing and context more important. The question is not only which model is ahead, but also how durable that lead appears to be.
The GPT-4 period showed what a large lead can look like. The more recent period shows a field where leading models are challenged quickly, and where the distance between leaders may be narrower.
- GPT-4: about a year at the top of the ECI.
- OpenAI's o1: just over three months in first place.
- Since February 2024: 17 changes in the lead after Claude 3 Opus dethroned GPT-4.
- Current pattern: a median stay of about seven weeks at the top per model.
A faster race with smaller gaps
The clearest takeaway is that AI model dominance has become harder to sustain. GPT-4's long lead now stands out precisely because newer models have not repeated it.
The ECI trend points to a market where progress is still rapid, but the lead is more contested. Instead of one model defining the top tier for about a year, leadership now moves among competitors on a much shorter cycle.
That makes the race more dynamic and less predictable. It also makes the meaning of “best model” more time-sensitive than it was during GPT-4's long run.