Mixed user reception for OpenAI's reasoning model—benchmarks improve but real-world coding and hallucination issues persist.