Use-case preset
Long-form article writing cost calculator
Generate articles or reports from a brief; output-token-heavy workload.
Article or report generation from a brief: the prompt provides an outline, tone instructions, reference facts, and style guidelines; the model outputs 800–2000 words of structured prose. This is an output-heavy, offline workload — quality and coherence matter far more than speed.
The 30/70 input/output split flips the typical ratio — the brief is concise, the article is long. An 8k context window comfortably holds the prompt plus a 1500-word output in a single pass. Latency is best-effort; writers queue jobs and review results asynchronously. `cachedPromptPercent` is low (~10) because the brief changes every job. The primary cost variable is output token count — a 2000-word article at average 1.3 tokens/word is ~2600 output tokens, so a 70B model at $1/1M output tokens costs ~$0.003 per article. Quality uplift from larger models is real here; thin output from a weak model wastes the entire human review step.