Vibe Check: Anthropic Just Made Opus Cheaper Without Calling It That

Katie Parrott’s day-zero review of Sonnet 4.6 finds it delivers Opus-close intelligence at roughly half the API cost ($3/$15 per million tokens vs. Opus pricing). The Reach Test yielded yellow ratings from both Dan Shipper and Kieran Klaassen — strong for production apps, not quite compelling enough to replace Opus for personal complex work.

Key findings: the model handles multi-step coding tasks, PR triage, and P&L restructuring without the mid-task mistakes of Sonnet 4.5. But speed, historically Sonnet’s advantage, is no longer meaningfully faster than Opus. Under pressure it showed erratic behavior — simultaneously too cautious and too eager. Spiral’s token costs (up to $1,000/day on Opus) would roughly halve without code changes.

Shipper’s strategic read: Anthropic keeps tier prices stable while improving capability. Whatever Opus does today, Sonnet will do for less in six months.

RDCO mapping: Cost-optimization signal for our API usage. The “Sonnet catches up in six months” pattern means we should design our model routing to easily swap tiers. The erratic-under-pressure finding is relevant to our autonomous loop reliability.