Pricing
Pay for the lane you use. Never for the wait.
All rates are per 1,000 tokens, input + output combined, on your dedicated lane. No per-request minimums. No overage surprise — overflow is metered, tagged, and visible in your dashboard.
Rate card
Token rates
| Tier | Rate (per 1K tokens) | Lane | Typical user | Notes |
|---|---|---|---|---|
| T1 | $0.003 | TinyLlama / Phi-3 class general containers | Prototypes, batch jobs | Basic + Query agents only |
| T2 | $0.007 | Small/mid models | Indie devs, small teams | Adds C4 access |
| T3 | $0.015 | Mid/large models (Gemma-4 12B class) | Prosumers, small teams | ResearchAgent included |
| T4 | $0.030 | Specialized multimodal lane | Mid-size teams, product teams | Vision, Code, Ops agents. Approval gate |
| T5 | $0.060 | Ultra-premium multimodal lane | Enterprises, high-volume teams | Orchestrator + Strategy agents. Approval gate |
Agent surcharges
Agents cost more than chat — so we price them honestly
Agent calls burn more tokens than single-turn chat. On T1–T2 lanes the agent surcharge keeps the lane affordable for everyone. From T3 up it's included in the rate.
| Tier | Light agent | Heavy agent | Extreme (long-running) |
|---|---|---|---|
| T1 | $0.004 / 1K | $0.004 / 1K | $0.010 / 1K |
| T2 | $0.004 / 1K | $0.004 / 1K | $0.010 / 1K |
| T3 | Included | ||
| T4 | Included | ||
| T5 | Included | ||
Estimate
What would your workload cost?
Enter your average monthly token volume, pick your tier, and set your agent share. The math below uses the published rate card — no hidden multipliers.
5,000,000 tokens ≈ 3.8M words
Estimated monthly cost
- Base lane tokens
- $0.00
- Agent surcharge
- $0.00
- Cloud overflow (2.5× tier rate)
- $0.00
Overflow
When your lane saturates
Your lane is reserved, but it isn't infinite. When load exceeds your lane's capacity, RabbitOS3 routes the excess to cloud endpoints automatically. Overflow is metered at 2.5× your tier rate, tagged on every request, and never blocks your traffic.
| Tier | Overflow rate (per 1K tokens) | Overflow destination | Notes |
|---|---|---|---|
| T1 | $0.0075 | Novita cloud endpoints | 2.5× base |
| T2 | $0.0175 | Novita cloud endpoints | 2.5× base |
| T3 | $0.0375 | Novita cloud endpoints | 2.5× base |
| T4 | $0.075 | Novita cloud endpoints | Reserved endpoints first, 2.5× base |
| T5 | $0.150 | Novita cloud endpoints | Reserved endpoints first, 2.5× base |
Sizing rule: price the peak, not the average. If your traffic is bursty, size your tier for the peak and treat only the genuinely spillover volume as overflow — that's almost always cheaper than running a top-tier lane all month.
Monthly bands
Commit less, or commit more
Bands are suggested monthly spend ranges. The floor is the minimum monthly commitment; the top band caps what we'll provision for you on that tier.†
| Tier | Floor (monthly min) | Mid band | Top band | Best for |
|---|---|---|---|---|
| T1 | $5/mo | $25–100/mo | $500/mo | Prototypes, batch jobs |
| T2 | $10/mo | $50–200/mo | $1,000/mo | Indie devs, small teams |
| T3 | $25/mo | $100–500/mo | $5,000/mo | Prosumers, small teams |
| T4 | $50/mo | $250–1,000/mo | $25,000/mo | Mid-size teams, product teams |
| T5 | $100/mo | $500–5,000/mo | $100,000+/mo | Enterprises, high-volume teams |
† Proposed values — final terms confirmed at onboarding.
Guarantees
SLO guarantees
Every tier comes with a live SLO/SLA status visible in your Control Tower dashboard. Availability is measured monthly; latency is p95 first-token on the lane.
| Tier | Monthly availability | p95 first-token | Support |
|---|---|---|---|
| T1 | 99.5%† | ≤ 2.0s† | Community Discord |
| T2 | 99.5%† | ≤ 1.5s† | Community Discord + email |
| T3 | 99.9%† | ≤ 1.2s† | Email support (1 business day) |
| T4 | 99.9%† | ≤ 1.5s† | Dedicated support channel |
| T5 | 99.9%† | ≤ 1.5s† | Priority + dedicated channel |
What happens when we miss
For each 0.1% below the availability target, you get a 10% credit on that month's lane bill per missed target, capped at 50% of the monthly total.† Credits apply automatically — no claim form, no support ticket required.
† Proposed targets — confirmed at onboarding.
Feature matrix
What's in each tier
| Feature | T1 | T2 | T3 | T4 | T5 |
|---|---|---|---|---|---|
| Dedicated inference lane | ✓ | ✓ | ✓ | ✓ | ✓ |
| Model class | TinyLlama / Phi-3 | Small/mid | Gemma-4 12B class | Specialized multimodal | Ultra-premium multimodal |
| Context window | 8K† | 16K† | 32K† | 64K† | 128K† |
| Agents — Basic + Query | ✓ | ✓ | ✓ | ✓ | ✓ |
| Agents — DataAgent | — | ✓ | ✓ | ✓ | ✓ |
| Agents — ResearchAgent | — | — | ✓ | ✓ | ✓ |
| Agents — Vision / Code / Ops | — | — | — | ✓ | ✓ |
| Agents — Orchestrator + Strategy | — | — | — | — | ✓ |
| Agent surcharge | $0.004–0.010/1K | $0.004–0.010/1K | Included | Included | Included |
| Cloud overflow (auto) | ✓ | ✓ | ✓ | ✓ (reserved first) | ✓ (reserved first) |
| Live SLO/SLA dashboard | ✓ | ✓ | ✓ | ✓ | ✓ |
| Approval gate | — | — | — | ✓ | ✓ |
| Monthly floor† | $5 | $10 | $25 | $50 | $100 |
Sizing examples
Three real shapes of workload
Indie dev · chat product
T3 lane, 1.5M tokens/mo, ~15% agent volume.
Startup · agent-heavy pipeline
T4 lane, 8M tokens/mo, 60% agent volume, 1.2M overflow/mo.
At T5 the base would be $480 + $180 overflow = $660 — same overflow, double the bill. The lane-fit is the point.
Enterprise · bursty multimodal
T5 lane, 25M tokens/mo average, 40% agent volume, 8M overflow/mo.
Bursty workloads: if overflow regularly exceeds 25% of base volume, re-tier up — it's cheaper than paying the 2.5× overflow rate month after month.
FAQ
The questions, answered straight
What counts as a token?
Same definition as the OpenAI API: input + output combined. No hidden tokenization differences between our endpoint and what your SDK reports.
Is the agent surcharge per-request or per-token?
Per-token, applied only to tokens consumed by agent calls on T1–T2 lanes. Light and heavy agents share the same surcharge rate; long-running orchestration (extreme) is billed at the extreme rate. From T3 up, agent surcharges are included.
How does overflow work?
When your lane saturates, RabbitOS3 routes the excess to cloud endpoints automatically. You pay the tier's overflow rate (2.5× base) only for overflowed tokens. Every overflowed request is tagged in your dashboard so you can see exactly when and why it happened.
Do I pay for overflow even if my lane isn't full?
No. Overflow only occurs when your lane's capacity is exceeded. If your lane is sized for your peak, you'll rarely — or never — overflow.
What are the monthly bands?
The floor is the minimum monthly commitment for your tier; the top band is the maximum monthly spend we'll provision on that tier. If you're between bands, you're fine — bands guide provisioning, not billing. You're billed on actual token consumption within the band.†
What happens if you miss the SLO?
For each 0.1% below the monthly availability target, you receive a 10% credit on that month's lane bill per missed target, capped at 50% of the monthly total.† Credits apply automatically.
What's the difference between SLO and SLA?
The SLO is the target we engineer to (availability + latency). The SLA is the contractual consequence if we miss it (the credit above). Both are visible in your dashboard in real time.
How do I get to T4 or T5?
T4 and T5 sit behind an approval gate because the specialized and ultra-premium lanes are finite. Apply through the request access form with your workload details; we match lane-fit and reply within one business day.