Recent releases of Claude Opus 4.6 have demonstrated strong performance on Humanity’s Last Exam, a demanding benchmark of expert-level questions across science, mathematics, and humanities, with leading variants scoring in the mid-30s to low-40s percent range. This places Anthropic’s models competitively against Gemini 3.1 Pro Preview and GPT-5 series entries on the public leaderboards. Traders are watching for further gains from tool-augmented reasoning, expanded context handling, or minor version updates before the June 30 deadline, as incremental capability improvements in large language models have historically lifted benchmark results quickly. Any new official Anthropic announcement on enhanced thinking modes or training refinements could shift outcomes in the final weeks.