Google’s recent Gemini 4 series releases, including Argon and Carbon, have driven trader sentiment by posting leading results on refined Humanity’s Last Exam subsets such as the 1,000-question HLE-Diamond benchmark, where Carbon reached 69.1% in late October 2026. This has lifted implied odds for a 60%+ official score above 50%, though the market resolves solely on the full ~2,500-question HLE leaderboard at agi.safe.ai, where current Gemini variants trail Claude Opus 5.5 and Fable models at around 57% versus 61%. The gap reflects differences between closed-ended reasoning partitions on subsets and the broader expert-level knowledge demands of the complete benchmark. Further Gemini updates or configuration gains before year-end remain the key swing factor amid rapid frontier model iteration.