Testing harness. Runs the real workflow against live OpenRouter for the demo brand “Hydra” — plan → copy variants → deterministic validation → cross-model critic.
"Fits Any Desk" Proof-First · meta_feed · ← new brief
| tier | model served | in | out | reasoning | $ |
|---|---|---|---|---|---|
| quality | anthropic/claude-sonnet-5 | 507 | 927 | 198 | $0.010284 |
| bulk | deepseek/deepseek-v4-pro | 417 | 48 | 0 | $0.000382 |
| bulk | deepseek/deepseek-v4-pro | 219 | 59 | 0 | $0.000586 |
| bulk | deepseek/deepseek-v4-pro | 221 | 47 | 0 | $0.000446 |
| critic | google/gemini-3.1-pro-preview | 652 | 2986 | 2594 | $0.037136 |
| total (5 calls) | $0.048835 | ||||