What Prediction Markets Are Forecasting
Prediction markets currently estimate there is a very high probability, roughly 9 in 10, that an Anthropic Claude artificial intelligence model will score at least 35% on a test called "Humanity's Last Exam" by the end of June 2026. This shows that traders who follow AI progress collectively believe this milestone is almost certain to be achieved.
Why the Market Sees It This Way
The high confidence stems from the nature of the test and recent AI advances. "Humanity's Last Exam" is a public benchmark created by the company Scale AI. It is designed to be extremely difficult, testing an AI's ability to reason through long, complex problems that often require knowledge of computer programming, mathematics, and logic. A score of 35%, while far from a perfect grade, is seen as a significant technical hurdle.
The market odds reflect two main factors. First, AI models from Anthropic and other labs like OpenAI have shown rapid improvement on similar reasoning benchmarks over the past two years. Each new model release tends to jump significantly in performance. Second, the deadline is still over two years away, which the market views as a long time for AI research to progress. Given the current pace, traders expect that by mid-2026, Claude models will have advanced enough to clear this 35% threshold.
Key Dates and Events to Watch
The most important signals will be new model releases from Anthropic. Each time Anthropic launches a new version of Claude, such as a hypothetical "Claude-4" or subsequent iteration, its score on the public Humanity's Last Exam leaderboard will be immediately updated. A large jump in score with a new release would solidify the current prediction, while a smaller-than-expected gain could cause the probability to drop.
Other events to watch include major releases from competitors like OpenAI's GPT-5 or Google's Gemini. Strong performance from those models on the same exam would suggest the overall field is advancing quickly, supporting the high probability for Claude. The market will likely become more volatile in the months leading up to the June 2026 deadline as the window for a necessary breakthrough narrows.
How Reliable Are These Predictions?
Prediction markets have a mixed but interesting record on long-term tech forecasts. They are often good at aggregating expert sentiment about technological trends, but they can be overly optimistic about the speed of development. For a concrete benchmark like this, where performance is publicly scored, markets tend to be more reliable than for vaguer questions. However, the two-year timeframe is a long one in AI. A major slowdown in research progress, or a strategic decision by Anthropic to focus on capabilities other than this specific test, could make the current 93% probability look too high in hindsight.
