Open-Source AI Is Closing the GPT-5 Gap. But How Fast?

Open models have closed most of the GPT-class gap on everyday agent work, but proprietary systems still win on hard reasoning, and free weights are not free to serve.

If you build agents for a living, you probably noticed two things this year. Frontier closed models keep getting better on paper. At the same time, more developers say the models they pay for feel worse in day-to-day use, while a smaller open model handles the boring 80% of their workload just fine.

So, how close are open-source models to GPT-class perfo…