Anthropic ships a faster, cheaper reasoning model

Anthropic’s latest release focuses on lowering the cost and latency of reasoning-heavy tasks, without giving up the accuracy gains of previous generations. For teams running agents in production, that mostly shows up as a lower bill for the same workload — not a new capability so much as a better price point for one that already existed.

Why this matters in practice

Reasoning-heavy calls — multi-step tool use, long-context analysis — have historically been the most expensive part of running an agent pipeline. A meaningful drop in per-token cost for that tier changes the math on which workflows are worth automating.