Effort level: medium

The cost-conscious production default

Medium effort is built for exactly the workloads that need real reasoning but also run at enough volume that cost genuinely matters — production pipelines processing meaningful traffic where the default, higher tier would work fine but wouldn't be the efficient choice at scale. On the current top-tier Sonnet generation, this tier lands roughly comparable to the previous generation's default tier, which makes it a natural landing point for a workload migrating forward that wants to hold cost roughly steady rather than accept the higher tier's cost increase by default.

Where it under-delivers

The known risk here is scope: this tier can occasionally handle an open-ended task more narrowly than the task actually called for, producing a correct but incomplete answer rather than a wrong one. That's a different failure shape than the lowest tier's under-thinking — the reasoning is real, just sometimes narrower in scope than an unconstrained task needed.

Testing before committing a workload here

Because this tier sits in the middle of the range specifically as a cost-quality compromise, it's worth validating directly against your own task distribution rather than assumed to be a safe universal choice — a workload with more open-ended tasks than typical may need the default tier above it more often than this one alone can reliably cover.

Related

See effort level: high for the tier this one is most often weighed against, and choosing effort level for cost, not just quality for the broader cost-segmentation strategy this tier fits into.