Operational Transparency
Fleet Cost Dashboard
CUBEcommons builds with AI agents. This page shows what that costs.
STU Normalization Table
Standard Token Unit — cost-normalized comparison across models. Baseline: Sonnet = 1.0×.
| Model | Multiplier | Verified | Source |
|---|---|---|---|
| Opus | 5.0× | 2026-04-07 | Anthropic pricing |
| Sonnet | 1.0× (baseline) | 2026-04-07 | Baseline |
| Haiku | 0.27× | 2026-04-07 | Anthropic pricing |
| GPT-4o | 1.2× | — | Approximate estimate |
| Local | 0.0× (configurable) | — | Infrastructure cost only |
STU multipliers are cost-weighted, not capability-weighted. Dates >90 days old will show an amber stale warning.
Monthly Fleet Burn Rate
Live data pending cost tracking pipeline deployment.
Monthly burn (STU)
—
Monthly burn (USD)
—
Monthly sessions
—
Model Tier Breakdown
100% Opus (current)
Will diversify when fleet adds Sonnet/Haiku agents for mechanical and search tasks.
Proving Ground Experiment Costs
| Cadence | Sessions/month | API Cost/month | Infra/month | Total/month |
|---|---|---|---|---|
| Light (1/week) | ~32 | $75–$450 | ~$4 | $79–$454 |
| Medium (1/day) | ~120 | $283–$1,674 | ~$14 | $297–$1,688 |
| Heavy (continuous) | ~480 | $1,133–$6,696 | ~$56 | $1,189–$6,752 |
API token costs dominate. Infrastructure cost is noise (~0.3% of total).
Reference Values
- Cost per session (Sonnet)
- $2.36–$13.95
- Cost per 48h experiment
- $190–$390
- Optimal session duration
- 45–60 min
- Kill threshold
- 70% context or 60 min
Cost is infrastructure. Transparency is architecture.