Trending on X: Claude Opus 4.6 "Nerfed" Debate Splits Power Users and Anthropic Staff
What Happened
AMD's Stella Laurenzo analyzed 6,852 Claude Code sessions and found median thinking block length fell from about 2,200 characters in January to 600 by March. A viral BridgeBench post claimed Opus 4.6 dropped from second to tenth on its hallucination leaderboard, though critics noted the expanded test set invalidated the comparison. Anthropic said the change reflects a shift to adaptive thinking defaults, not model modification.
My Take
Power users are treating reasoning token length as a direct proxy for model intelligence — a fragile assumption. What Anthropic actually did was shift defaults to reduce compute per query, a rational business move when infrastructure costs are scaling alongside $30 billion in annual revenue. Every frontier lab will face the same pressure and adopt tiered reasoning budgets within six months. The loudest complainers will get a premium compute toggle. The majority of users won't notice or care.
Read Original Source