Artificial Analysis Says Grok 4.6 Finishes Long-Horizon Tasks in 53 Turns vs Opus 5's 103 — and a Quarter of the Input Tokens
Artificial Analysis / Hacker News·high signal
The independent Artificial Analysis teardown drew its own HN thread (334 points, 381 comments) largely because of one number: on long-horizon knowledge work, Grok 4.6 resolves tasks in ~53 turns and ~0.5B input tokens against Claude Opus 5's ~103 turns and ~2.0B, landing a measured cost-per-task of $0.84. Grok 4.6 still scores below Opus 5 (63) and Fable 5 (62) at 61. For anyone paying per token rather than per seat, turn efficiency — not index score — is the line item that moves, and this is the first head-to-head with turn counts attached.