Google’s Gemini series has shown strong but not always leading performance on Humanity’s Last Exam, a demanding benchmark of expert-level reasoning across academic fields. Gemini 3 Pro set an early record of 37.5% upon its November 2025 launch, followed by Gemini 3.1 Pro reaching around 44-46% and later variants like 3.8 Flash and the newly released Gemini 4 Argon (October 2026) posting competitive no-tools scores near 44-45%. Anthropic’s Claude models currently top leaderboards at 60%+, reflecting faster progress by rivals in frontier reasoning. Google has canceled plans for a Gemini 3.5 Pro after delays, shifting focus to iterative Flash and specialized Deep Think releases. Traders watch for the next Pro-tier flagship’s benchmark results and any confirmed HLE evaluations, as these directly influence market-implied odds amid intense AI lab competition.