Highest Google Gemini score on Humanity’s Last Exam in 2026?
Google Gemini is still below 50% on the relevant Humanity’s Last Exam benchmark variants, but there is still enough time in 2026 for a new release or updated evaluation to clear the threshold. I think the market is a bit too confident, yet the underlying trend still makes Yes more likely than not.
Analysis
The strongest evidence available today points to Gemini still missing the 50% bar on the benchmark that matters for this market. The latest reported results put Gemini 3.8 Flash at 34.3% on HLE-Diamond, 42.1% on an AA text-only HLE entry, and an older Gemini 3.1 Pro result at 44.4% without tools on Humanity’s Last Exam. Those are all materially below the threshold, which means the market is not currently being supported by an already-observed score at or above 50% on the exact target benchmark. In other words, the Yes case still depends on future progress rather than current evidence.
At the same time, the market is asking about the highest Gemini score achieved during all of 2026, not just the latest public score. That matters because benchmark progress in frontier models can move quickly, and Google has strong incentives to push Gemini on highly visible reasoning benchmarks before the end of the year. A jump from the low-to-mid 40s to 50% is not trivial, but it is also not absurd if a newer model, a larger compute setup, improved tool use, or a refined benchmark-specific adaptation is released before year-end. The fact that Gemini has already moved into the 40s on a closely related setup suggests there is a plausible path to crossing 50%, even if it has not happened yet.
Against Yes, the recent pattern is not especially encouraging because the newest cited Gemini numbers are still below 50% and, in one case, substantially below it on a variant benchmark. The gap to 50% is small but persistent, and benchmark variants matter here because the market resolves only on the official Humanity’s Last Exam leaderboard or a clear equivalent metric. If Google’s strongest public 2026 result remains in the low-to-mid 40s, this market should resolve No. But because there are several months left in the year and the frontier model race is intense, I still lean that a breakthrough score is more likely than not, just not nearly as certain as the current market price implies.
Arguments
For
- Arguments for Yes: Google has shown steady benchmark improvement, and a move from the mid-40s to at least 50% is plausible within a frontier model cycle.
- Arguments for Yes: The market only needs one qualifying Gemini result in 2026, so a late-year release or updated leaderboard entry could settle the question in favor of Yes.
Against
- Arguments against Yes: The most relevant published Gemini scores are still 34.3%, 42.1%, and 44.4%, which indicates a real gap to 50%.
- Arguments against Yes: If the next Gemini release improves more slowly than expected or is evaluated on a slightly different HLE variant, the official score may still fall short.
Key drivers
- Gemini’s latest cited scores remain below 50%, so the outcome depends on future model improvements rather than current performance.
- There is still enough of 2026 left for Google to release a stronger Gemini model or a revised evaluation run that clears the threshold.
Risk factors
- The official Humanity’s Last Exam metric may continue to lag behind the threshold even if Gemini improves on related benchmarks.
- Google may prioritize broader product quality over a benchmark-targeted push, leaving the best 2026 score below 50%.
Scenarios
Best case
Google releases a materially stronger Gemini model in 2026 and it posts at least 50% on the official Humanity’s Last Exam leaderboard, making the market resolve Yes.
Most likely
Gemini continues improving and may get close to 50%, but the decisive official score remains uncertain until a late-2026 release or leaderboard update, leaving Yes somewhat favored but far from guaranteed.
Worst case
Gemini’s 2026 best result tops out in the mid-40s on the official benchmark, with related variants improving but never crossing 50%, so the market resolves No.
More from this day
- SportsKalshi3mo
When will Greg Schiano leave Rutgers?
AI28%MKT93%Edge-65HypedI think Greg Schiano is much more likely to remain Rutgers coach through Feb. 1, 2027 than the market suggests. The current record has created pressure, but there is no concrete departure signal yet, and the price appears to be extrapolating hot-seat chatter far beyond the evidence.
- PoliticsKalshi20d
Will 3 mg THC beverages qualify as hemp under federal law?
AI28%MKT92%Edge-64HypedMy estimate is that 3 mg THC beverages will probably not qualify as hemp under federal law on Nov. 13, 2026. The reported container-based THC cap is far below 3 mg, and the main uncertainty is timing rather than substance.
- techPolymarket2mo
How many more Millennium Prize Problems will AI solve in 2026?
AI28%MKT69%Edge-41HypedI think there is a low chance that AI ends 2026 having solved zero additional Millennium Prize Problems, because there is already a public claim from an eligible lab that it solved Navier-Stokes. The main uncertainty is whether that claim ultimately counts under the market’s exact wording, but absent a major reversal, the zero-outcome looks weaker than the market price implies.