Testing You.com's API for ad copy generation. Need to know which models are cost-effective for high-volume, repetitive tasks.
Ran 100 sequential prompts through the playground, tracking latency and token cost. Tested `you-codellama-34b`, `you-mistral-7b`, and the default `you-chat` model.
- **you-codellama-34b**: Highest quality output, but 2.1s average latency. Token cost is 3x the 7B model.
- **you-mistral-7b**: Fastest (0.8s avg). Output is formulaic but usable for simple variants. Lowest cost.
- **you-chat**: Unpredictable latency spikes (0.5s to 4s). Inconsistent output formatting.
For bulk generation of simple headline variants, `you-mistral-7b` is the only viable option. The others don't scale. Need to see if the API allows persistent model selection for automation, or if it defaults. Documentation is unclear.