Cost & Efficiency
Token Margin Tracker
Estimate the gap between public token prices and rough raw hosting costs for open-weight model serving.
How to use this dashboard
Estimate the gap between public token prices and rough raw hosting costs for open-weight model serving.
Use this tracker to estimate the spread between raw hosting assumptions and public API token prices. Treat it as a pricing signal, not a profit-margin claim.
Token Margin Tracker
11 records| API wrapper / router | Open-weight 8B chat model | 0.18 | 0.04 | 4.5x | High utilization on low-cost GPU; excludes engineering, support, margin, failed generations, and idle time | Low / estimate |
| Hosted open model API | Open-weight 70B chat model | 0.9 | 0.28 | 3.2x | Steady batch traffic on marketplace GPU; excludes redundancy and orchestration cost | Low / estimate |
| Premium proprietary API | Frontier reasoning model | 15 | Unknown | Not knowable | Closed model. Raw cost cannot be verified from public data. | Unknown |
| GPU marketplace self-host | Self-hosted embedding model | 0.1 | 0.015 | 6.7x | Embeddings are easy to batch; utilization matters more than peak GPU speed | Medium estimate |
| Serverless inference provider | Open-weight mixture-of-experts model | 0.65 | 0.18 | 3.6x | MoE serving has routing/VRAM complexity; raw GPU cost is not the full cost | Low / estimate |
| OpenRouter | Public model and pricing catalog | 0 | Not knowable from public data | Unknown | Closed/private cost stack. Public API price can be shown, but true raw serving cost cannot be verified. | Public price only |
| Openrouter | Pareto Code Router | -1,000,000 | Not knowable from public data | Unknown | Closed/private cost stack. Public API price can be shown, but true raw serving cost cannot be verified. | Public price only |
| Openrouter | Body Builder (beta) | -1,000,000 | Not knowable from public data | Unknown | Closed/private cost stack. Public API price can be shown, but true raw serving cost cannot be verified. | Public price only |
| Nvidia | NVIDIA: Nemotron 3 Nano Omni (free) | 0 | Not knowable from public data | Unknown | Closed/private cost stack. Public API price can be shown, but true raw serving cost cannot be verified. | Public price only |
| Poolside | Poolside: Laguna XS.2 (free) | 0 | Not knowable from public data | Unknown | Closed/private cost stack. Public API price can be shown, but true raw serving cost cannot be verified. | Public price only |
| Poolside | Poolside: Laguna M.1 (free) | 0 | Not knowable from public data | Unknown | Closed/private cost stack. Public API price can be shown, but true raw serving cost cannot be verified. | Public price only |