MiraBridge uses two clear limits on paid usage:
- Weighted fast requests: priority runs that execute immediately until your plan quota is exhausted.
- Standard monthly token budget: the included compute budget for the billing period. Higher-cost models consume this budget faster.
Every plan can access published MiraBridge models while fast quota is available. We do not hide models behind separate add-ons. Instead, higher-cost models consume more fast-request credits and more standard token budget, while the slow queue is reserved for lighter managed models after fast quota is exhausted.
Plan Comparison
| Feature | Free | Pro | Pro+ | Ultra | | --------------------------- | ---------- | ------------- | ------------ | ------------ | | Weighted fast requests / mo | 50 | 150 | 280 | 900 | | Standard token budget | 50K | 1M | 2M | 6M | | Models | Fast quota | Fast quota | Fast quota | Fast quota | | Lighter-model slow queue | No | Yes | Yes | Yes | | Approx. Opus fast runs / mo | 3 | 10 | 18 | 60 | | Devices | 2 | 3 | 5 | 10 | | Tool approval | Manual | Manual + Auto | Batch + Auto | Batch + Auto | | Support | Community | Email | Priority | Dedicated |
How Request Weights Work
Fast request weights are coarse fairness bands. The same per-model weight is used both to enforce the priority quota and to draw down the standard token budget, so expensive models are available without making the plan economically unsafe.
Request weights change as the model catalog changes: new models arrive, some are promoted from preview to GA, and others are deprecated. The single source of truth is the live pricing page:
- mirabridge.io/pricing â current model availability, per-model request weights, and active provider list. The gateway uses the same catalog for checkout and usage accounting, so what you see on
/pricingis exactly what gets charged.
As a rule of thumb:
- Lightweight models (Gemini Flash Lite / Claude Haiku tier) sit at the cheap end of the weight scale.
- Mid-tier flagships (Claude Sonnet, mid-range GPT-5, Gemini 2.5 Pro) sit in the middle.
- Premium reasoning models (Claude Opus, GPT-5.5, o-series) sit at the high end.
Calibrate against your current spend on /pricing rather than copying static numbers from this page.
BYOK Alternative
With Bring Your Own Key, you pay the AI provider directly and bypass MiraBridge request-credit billing entirely. BYOK is available on paid plans.
Request Packs
If you need more priority capacity, you can buy request packs. Request packs add extra fast-request credits and do not expire. They are best for bursts, while the standard token budget still protects overall compute costs.
Next Steps
- Enterprise â team controls, BYOK routing, and audit-friendly workflow
- BYOK â use your own API keys