OpenAI's GPT-5 series maintains a commanding lead on FrontierMath benchmarks and open-problem attributions, with models like GPT-5.6 Sol topping the v2 Tier 4 leaderboard at 83% as of late August 2026 and the lab credited with the bulk of roughly 234 solved problems overall. This stems from iterative scaling of reasoning agents, heavy test-time compute, and early funding ties to the benchmark itself. Anthropic's Claude Fable 5 and related models delivered the second and third verified solutions to genuine unsolved problems in July and mid-August, including a 40-year-old number theory result and a Hadamard matrix construction, narrowing the gap on the 50-problem Open Problems track. Google DeepMind's multi-agent systems also post competitive Tier 4 scores. Traders are watching for the next verified AI solution amid ongoing benchmark expansions, as additional problems could quickly shift implied probabilities toward whichever lab demonstrates the next breakthrough.