Highest Claude score on Humanity’s Last Exam in 2026?

The central fact is that Claude is already very near the threshold. The most credible public figures in the current context place Anthropic's best official or near-official Humanity's Last Exam result around 53.3%, with another cited result at 52.6%, so only a small additional improvement is needed to reach 55%. That small gap matters because for frontier models, a one- to two-point gain on a hard benchmark can arrive with a new model revision, a better inference setup, or a more optimized evaluation run.

At the same time, the benchmark reporting landscape is messy, and that cuts both ways. Some trackers report higher Claude scores, including values above 55% when tools are allowed or under slightly different conditions, but the market resolves on the official Humanity's Last Exam leaderboard and on the HLE Accuracy measure specifically. If the public leaderboard is the only thing that counts, then a score that looks strong in an alternate setup may not help unless Anthropic posts a comparable official result that clearly clears the threshold.

The time remaining is enough for another attempt, but not enough to assume one. Anthropic has strong incentives to keep pushing frontier performance, and a year-end release could plausibly add the needed couple of points. Still, the market price implies a much higher chance than the raw official evidence alone suggests, so I think traders are leaning heavily on the possibility of a late-year leap or on already-available noncomparable results; my independent read is that Yes is favored, but not nearly as strongly as the current price indicates.