Highest score on Humanity’s Last Exam in 2026?
This event is a set of four markets. Each asks whether any AI model will reach at least a given accuracy (70%, 75%, 80% or 90%) on Humanity's Last Exam by the end of 2026. The deciding number is the "HLE Accuracy" column on the official leaderboard at agi.safe.ai. Calibration Error and other columns do not count.
Leading right now: 70%+ 28%, 75%+ 11%, 90%+ 7%, 80%+ 7%.
| Market | Chance of yes | Deadline |
|---|---|---|
| Will the highest score achieved on Humanity’s Last Exam in 2026 be 70% or higher? | 28.5% | Jan 1, 2027, 4:59 UTC |
| Will the highest score achieved on Humanity’s Last Exam in 2026 be 75% or higher? | 11.4% | Jan 1, 2027, 4:59 UTC |
| Will the highest score achieved on Humanity’s Last Exam in 2026 be 90% or higher? | 7.1% | Jan 1, 2027, 4:59 UTC |
| Will the highest score achieved on Humanity’s Last Exam in 2026 be 80% or higher? | 7.0% | Jan 1, 2027, 4:59 UTC |
Odds read from Polymarket at 21:54 UTC on Nov 21, 2026. $189K traded so far, $4K in the last 24 hours.
How this market resolves
Yes: For a given market, YES requires that any model has an "HLE Accuracy" value at or above that market's threshold (70%, 75%, 80% or 90%) on the official leaderboard by December 31, 2026, 11:59 PM ET. A clearly equivalent metric also counts if the leaderboard is restructured or renamed.
No: A market resolves NO if no model reaches its threshold on that metric by the deadline. It also resolves NO if the leaderboard is unavailable after the deadline or confirmed permanently gone, and no official alternative source exists.
When it really ends
December 31, 2026, 11:59 PM ET (Eastern Time, UTC-5 on that date), which is 2027-01-01 04:59 UTC. The endDate field in the data (2027-01-01T04:59:00Z) is the same moment, so there is no difference.
What counts
- Any model listed on the official leaderboard at agi.safe.ai, from any developer.
- The value labeled "HLE Accuracy", or a clear equivalent if the leaderboard is redesigned or relabeled.
- A score exactly equal to the threshold, since the rules say "at least".
- Official Humanity's Last Exam results published elsewhere, but only if the leaderboard stays unavailable after the deadline or is confirmed permanently gone.
What does not count
- Calibration Error or any other displayed metric.
- Scores reported only in a company's blog post, press release or social media post, unless they appear in the official results.
- Anything that appears on the leaderboard after the December 31, 2026, 11:59 PM ET deadline.
- Unofficial or third-party leaderboards while the official one is available.
Fine print people miss
- The titles say "in 2026", but the rules only require that a model reach the accuracy "by" the deadline. They do not say the score must first be set during 2026.
- The rules do not say how to treat different evaluation setups, such as models with tools versus without, or text-only subsets. The leaderboard's own presentation of "HLE Accuracy" is what governs.
- If the leaderboard goes offline during the period, the markets stay open until the data is available again.
- If the source is still unavailable after the deadline or confirmed permanently gone, other official HLE results may be used. If no official alternative exists, all markets resolve NO.
- The four markets are separate. A single top score can resolve some to YES and others to NO.
- The rules do not say how often the leaderboard is updated, so a score reached near the deadline may not appear in time.
What moves the price
- A new frontier model release with a higher reported HLE accuracy would raise the odds of the higher-threshold markets.
- A leaderboard update that adds a new top-scoring model would raise the odds for any threshold it reaches.
- Evidence that progress is slowing, or that new releases are not being added to the leaderboard, would lower the odds, especially for the 80% and 90% markets.
- Changes to how the leaderboard is run or presented, such as re-scoring, corrected answers or a new version of the exam, could move scores in either direction.
- The leaderboard going offline or being left stale near year end would lower the odds.
Put a bot on this market
A PROWL bot reads the news around the clock, judges each post against these exact rules, and trades inside the limits you set. It starts in paper mode.
Questions people ask
What is Humanity's Last Exam?
It is an AI benchmark with a public leaderboard at agi.safe.ai. These markets use the leaderboard's "HLE Accuracy" column to measure how well models do.
When do the Humanity's Last Exam markets end?
They end on December 31, 2026 at 11:59 PM ET. That is 2027-01-01 04:59 UTC. The listed end date matches the rules.
Which number on the leaderboard decides the market?
Only the value labeled "HLE Accuracy", or a clear equivalent if the page is restructured. Calibration Error and other metrics are ignored.
Does a lab's own announced score count?
Not by itself. The resolution source is the official leaderboard at agi.safe.ai. Other official HLE results are used only if that site is unavailable after the deadline or permanently gone.
What happens if the leaderboard website goes down?
The markets stay open until the data is available again. If it is still down after the deadline or confirmed permanently gone, other official HLE results may be used. If none exist, the markets resolve NO.
Can more than one of the markets resolve YES?
Yes. Each market has its own threshold. If a model reaches a high threshold, it also reaches every lower one, so those markets resolve YES too.
More AI markets explained
Written from the market's own rules text on Nov 21, 2026. If the rules on Polymarket differ from this page, the rules on Polymarket govern.