OpenAI’s GPT models have shown steady gains on Humanity’s Last Exam, a 2,500-question benchmark spanning expert-level topics in science, math, and the humanities, with the latest GPT-5.4 and GPT-5.5 variants posting scores between 41% and 58% as of mid-May 2026. This marks substantial progress from the 2–8% range recorded by GPT-4o and o1 when the benchmark launched in early 2025. Iterative releases, including higher-compute “Pro” and “Thinking” variants, along with internal scaling of training and inference, have driven these lifts, though Google’s Gemini 3.1 currently leads the overall leaderboard. Traders are watching for any June model update or fine-tune that could push an OpenAI entry past the current frontier before the June 30 cutoff.