Opus 4.6 Thinking keeps the #1 spot. Followed by Opus 4.7 Thinking (-15 points). Lastly, Opus 4.8 Thinking (-23 points compared to 4.6 Thinking). https://arena.ai/leaderboard/text/hard-prompts-english As a non-coder, I find the Hard Prompts (English) benchmark on LMArena to be the one that best matches my experience at work. It's probably more immune to benchmaxxing. Simple Bench also shows that 4.6 is the best model in the Opus family. submitted by /u/Legitimate...