What will be a model's highest score as of 18 December 2026 on Humanity's Last Exam (HLE), according to the Center for AI Safety (CAIS)?
Closing Dec 18, 2026 08:01AM UTC
HLE is "a multi-modal benchmark at the frontier of human knowledge, designed to be the final closed-ended academic benchmark of its kind with broad subject coverage" with 2,500 questions spanning over 100 subjects (CAIS - HLE). The question will be suspended on 17 December 2026 and the outcome determined using the scores as reported by CAIS at approximately 5:00PM ET on 18 December 2026 (CAIS AI Dashboard, see "Text Capabilities Index" and click the header for the "Humanity's Last Exam" column). As of 2 February 2026, Google's "Gemini 3 Pro" model had the high score of 38.3%.
Confused? Check our FAQ or ask us for help. To learn more about Good Judgment and Superforecasting, click here.
To learn more about how you can become a Superforecaster, see here. For other posts from our Insights blog, click here.
| Possible Answer | Crowd Forecast | Change in last 24 hours |
|---|---|---|
| Less than 40.0% | 0% | 0% |
| At least 40.0%, but less than 45.0% | 0% | 0% |
| At least 45.0%, but less than 50.0% | 5.00% | +5.00% |
| At least 50.0%, but less than 55.0% | 32.00% | -2.00% |
| At least 55.0%, but less than 65.0% | 47.50% | +2.50% |
| At least 65.0%, but less than 75.0% | 15.50% | -5.50% |
| 75.0% or more | 0% | 0% |