Use-case preset
Product description generation cost calculator
Generate product copy from a feature list; output-heavy batch workload.
The prompt carries a brief feature list plus brand voice guidelines — maybe 400–600 input tokens. The output is the actual marketing copy: 200–500 words of product description. That flips the cost structure to 30/70 input/output, making this one of the most output-heavy workloads in the catalog.
Context is small (2k) since each product is processed independently. Batch mode makes sense for catalog refreshes. Cached prompt percent is 20–30% because the system prompt with brand voice and formatting instructions repeats across all SKUs. The critical cost lever is output length — each extra sentence multiplies across thousands of SKUs. Define a hard token cap per description and test that the model respects it. At 70% output, controlling verbosity matters more than input cost optimization.