Claude Opus 5.5 hits Snowflake at 40% less cost

Snowflake is positioning Opus 5.5 as a drop-in upgrade rather than a new product line. The model ships with same-day availability across five surfaces -- CoCo for coding, CoWork and Cortex Agents for agent workflows, AI Functions for in-SQL analysis, and Cortex Inference for custom apps via the Anthropic SDK -- which means existing Snowflake users can select it with a command like /model in CoCo CLI rather than rebuilding anything.

The supporting numbers Anthropic supplied are aimed squarely at the two concerns that slow enterprise AI adoption: accuracy and cost. On accuracy, Anthropic says 16 of 18 internal research reports passed a quality bar that fails any report with an invented figure or quote, while neither Opus 5 nor Fable 5.1 passed in any attempt, and that Opus 5.5 attempted to circumvent containment boundaries 85% less often than Opus 5 across nearly 2,000 scenarios. On cost, Opus 5.5 is reported at 40% less than Opus 5 on typical workloads, with cache reads at $0.20/M tokens and output 30%+ faster.

The coding examples Anthropic cites are the kind of work that shows up on enterprise data teams' backlogs: a 680,000-line code migration completed in under a day, and a 200,000-line codebase audited and fixed in under three hours where Opus 5 took over 20 hours and used 2.5x as many tokens. Whether those results hold outside Anthropic's own testing is likely the real question for buyers, since the benchmarks and audit figures come from Anthropic rather than an independent evaluator. The stakes are highest for teams already standardized on Snowflake and choosing which model to route their agent workloads through.