OpenAI's reasoning AI models are getting better, but their hallucinating isn't, according to benchmark results.

AI models are numerous and confusing to navigate, but the benchmarks used to measure their performance are also challenging.