Highest Google Gemini score on Humanity’s Last Exam in 2026?
Gemini is already very close to the 50% threshold on Humanity’s Last Exam, so a year-end breakthrough is plausible. Even so, the current best-known Gemini score is still 47%, which makes the market’s very high Yes price look somewhat aggressive relative to the evidence.
Analysis
The clearest anchor is that the highest publicly visible Gemini result in the provided snapshot is 47.0%, which leaves only a small gap to the 50% line. That is important because this market resolves on the official Humanity’s Last Exam leaderboard and specifically on HLE Accuracy, so a single new Gemini model or updated run could change the outcome quickly if Google decides to submit one before year-end. On the other hand, being close to the threshold is not the same as crossing it, and the market still depends on an actual official leaderboard update rather than general impressions of model quality.
The broader benchmark context is encouraging for a Yes outcome because frontier systems are already clearing 50% on some HLE presentations, showing that the threshold is technically reachable rather than a distant target. Google also has a strong incentive to keep pushing Gemini performance, and the company has a track record of iterative releases that can move scores by a few points over a relatively short period. If Google has a stronger Gemini variant in the pipeline, a 3-point improvement from the current best-known score is not a stretch, especially given how aggressively frontier labs optimize for benchmark leadership.
The main reason to stay below the market price is that 47% still leaves a meaningful margin of failure, and benchmark gains tend to get harder as models approach a harder ceiling. A model can improve across many other evaluations without necessarily clearing this specific threshold, and there is no guarantee that Google will publicly post a new HLE run before December 31. Because the market is already pricing Yes near 90%, the implied assumption is that a qualifying Gemini result is highly likely, but the evidence provided here supports confidence rather than near certainty, so I view the true probability as solidly above even money but materially below the current price.
Arguments
For
- Arguments for Yes: Gemini is already close enough to the threshold that a modest official improvement would flip the market.
- Arguments for Yes: Google has strong incentives to push frontier benchmark performance and showcase a headline-grabbing result.
- Arguments for Yes: The existence of other models above 50% suggests the target is within reach for a top-tier Gemini update.
Against
- Arguments against Yes: The best known Gemini score is still below 50%, so the market still needs a concrete official breakthrough.
- Arguments against Yes: Benchmark gains become harder near the top end, and a 3-point increase is not guaranteed.
- Arguments against Yes: If Google does not publish a qualifying HLE score before the deadline, the market resolves No regardless of informal performance.
Key drivers
- Gemini’s best known HLE score is already 47.0%, only three points below the resolution threshold.
- Google could release a newer Gemini model or updated leaderboard run before year-end.
- Competitor models crossing 50% indicates the benchmark barrier is achievable with current-generation systems.
- The market is likely extrapolating from Google’s rapid model iteration, which can create a real chance of a late-year jump.
Risk factors
- Google may not submit a new official HLE result in time for the market to resolve Yes.
- Improving from 47% to 50% on a hard benchmark may require a substantial capability jump, not a small tuning gain.
- The best public Gemini score could remain unchanged even if other benchmarks improve.
- Official leaderboard wording and timing matter, so a non-qualifying or differently labeled metric would not count.
Scenarios
Best case
Google releases a materially stronger Gemini model later in 2026 and the official HLE leaderboard shows a score at or above 50%, making the Yes outcome straightforward.
Most likely
Google improves Gemini enough to remain in the conversation and may get very near 50%, but whether it crosses the line depends on a specific year-end release or leaderboard update.
Worst case
Gemini tops out below 50% for the rest of the year or Google never posts a qualifying official run, so the market resolves No.
More from this day
- CompaniesKalshi1y
Starbucks total global stores in 2026
AI97%MKT9%Edge+88Hidden GemStarbucks looks very likely to clear 41,800 global stores in 2026. The latest reported base and management’s own net-new store guidance point to a year-end total comfortably above the threshold.
- cryptoPolymarket3mo
What price will Ethereum hit in 2026?
AI97%MKT19%Edge+78Hidden GemEthereum looks overwhelmingly likely to hit $3,000 by December 31, 2026, and the provided context even suggests it may already have done so this year. The main uncertainty is not market direction but whether the event will resolve cleanly on a recognized price print.
- PoliticsKalshi2y
Which agencies will Trump eliminate?
AI84%MKT29%Edge+55Hidden GemUSAID looks much more likely than not to count as eliminated during Trump’s term, because the administration has already dismantled its independent operations and shifted its functions into the State Department. The main uncertainty is whether the market requires formal legal abolition, but even under that standard the path toward Yes remains materially stronger than the current price implies.