TechnologyClaude Opus 5.5 launches with a claimed 40% cost cut
A user's cost breakdown says the benchmark wins came at max effort, while the advertised 40% savings versus Opus 5 were measured at default effort.
TLDR
A user's cost breakdown argues that Claude Opus 5.5's benchmark wins and advertised savings apply at different effort settings: max for the benchmarks, default for the 40% cost reduction versus Opus 5. Citing Artificial Analysis, the user says max-effort Opus 5.5 generated roughly three times the median output tokens—260M versus 88M—with output priced at $20 per million tokens. The breakdown also reports a 60% drop in cache-read pricing, to $0.20 per million tokens.
Another post relays Anthropic's claim that the model runs over 30% faster than Opus 5, alongside Vercel's warning that requests forcing a tool call or pinning a thinking budget return HTTP 400 errors.
