Highest Kimi score on Humanity’s Last Exam in 2026?
I think Yes is somewhat more likely than not, but the case is not strong enough to justify the market’s current optimism. The core reason is that there are credible reports of Kimi reaching the high 40s, yet the official-looking visible leaderboard evidence still appears to sit below the threshold.
Analysis
The main question is not whether Kimi can get close to 45%, but whether the highest score achieved by any Moonshot Kimi model in 2026 will be recorded in a way that clearly satisfies the market’s resolution rules. The recent evidence points in two different directions at once. On one hand, there are multiple references to Kimi K3 landing in the low-to-mid 40s, and one especially important report says Kimi K3 reached a 47% max score on Humanity’s Last Exam. If that figure is accepted as an official or clearly equivalent HLE accuracy result, then the Yes outcome would be satisfied already. On the other hand, the visible leaderboard-style numbers that are easier to verify seem to cluster below 45%, including a 22.2% figure on one tracked variant and several mid-40s references that stop just short of the cutoff.
The market is therefore mostly a resolution-risk problem rather than a pure model-capability problem. Kimi appears to be near the threshold and capable of crossing it in at least one reported setting, but the final outcome depends on whether the relevant score is posted on the official Humanity’s Last Exam leaderboard, whether it uses the same evaluation setup, and whether the score is presented as the model’s HLE Accuracy or an equivalent metric the market would accept. That uncertainty matters a lot because prediction markets often overprice outcomes that are plausible in principle but harder to verify under the exact resolution language. The current market price of about 72% implies strong confidence that an acceptable 45%+ result will be visible by year-end, but the evidence does not look that decisive to me.
Historically, benchmark reports often show a gap between a model’s best claimed result and the score that ends up being easiest to resolve against. A model may have a “max” configuration, ensembling, or a variant-specific setting that produces a higher number, while the canonical public leaderboard remains lower. That pattern seems especially relevant here because the recent commentary explicitly distinguishes between regular leaderboard figures below 45% and a separate 47% max claim. My read is that Yes has a real path to happening, but the probability is moderated by the possibility that the market’s exact scoring and source rules end up aligning with the lower, more conservative figures rather than the best-case reported number.
Arguments
For
- Arguments for Yes: A reported 47% max score indicates Kimi has already demonstrated performance above the threshold in at least one public account.
- Arguments for Yes: The benchmark trend suggests Kimi is operating near the cutoff, so incremental updates or a finalized official posting could easily push the resolved score over 45%.
Against
- Arguments against Yes: The most directly visible leaderboard-style evidence still shows Kimi below 45%, which is the cleanest basis for resolution.
- Arguments against Yes: The market may require a specific official HLE Accuracy figure, and the higher max claim may not match that exact scoring context.
Key drivers
- A reported Kimi K3 max score of 47% would directly clear the market threshold if it is accepted under the resolution rules.
- Several public benchmark views still place Kimi in the low-to-mid 40s or lower, creating meaningful ambiguity about whether 45% will be officially documented.
Risk factors
- The official leaderboard may remain below 45% even if a higher unofficial or variant-based score has been reported elsewhere.
- Resolution could depend on a specific HLE accuracy presentation that does not treat a cited max score or alternative benchmark summary as sufficient.
Scenarios
Best case
Moonshot publishes or is credited with an official HLE result at 45% or above, and the market resolves Yes without ambiguity because the score is clearly tied to the exact metric the rules specify.
Most likely
Kimi remains very close to the threshold, and the final outcome hinges on whether one accepted official or equivalent result above 45% appears before year-end; I think that happens slightly more often than not, but not comfortably.
Worst case
The highest verifiable official score stays in the low or mid-40s, while the 47% claim remains non-resolving, variant-specific, or otherwise not acceptable under the market rules, leading to No.
More from this day
- PoliticsKalshi20d
Will 3 mg THC beverages qualify as hemp under federal law?
AI9%MKT87%Edge-78HypedA 3 mg THC beverage is very unlikely to qualify as hemp under the federal changes taking effect in November 2026. My independent probability for a Yes outcome is 9%, far below the market price.
- PoliticsKalshi2mo
When will a catastrophic AI duty of care become U.S. law?
AI58%MKT8%Edge+50Hidden GemI put the odds of Yes meaningfully above the market because several state laws now create AI safety obligations that plausibly fit a catastrophic-risk duty of care before the deadline. The main reason not to go much higher is that the exact contract wording may be read narrowly as federal law only, which would leave the outcome much less likely.
- PoliticsKalshi2mo
When will the debt limit be increased?
AI68%MKT24%Edge+44Hidden GemI think the market is substantially underpricing this event. Based on current reporting and the usual congressional pattern, there is a better-than-even chance that some form of debt-limit increase becomes law before Jan. 1, 2027.