Pick a model, set the expected input/output token counts, and adjust the FX rate. During the design-partner phase, what you pay through SchneeAI equals the provider list price — no markup. The platform fee is under adjustment.
What the platform fee will cover
The same prompt-caching, batch API, and model-routing optimizations are available to you with or without SchneeAI — we do not pretend otherwise. The platform fee (under adjustment) pays for the layers most teams eventually build on top of a raw proxy:
| Layer | What it does | Build-it-yourself effort |
|---|---|---|
| LiteLLM routing + failover | One API for 100+ providers, fallback chains | 1–2 eng-weeks initial, ongoing maintenance |
| Vault (encrypted raw retention) | Audit-grade storage for prompts + outputs | 3–6 eng-weeks + KMS + legal review |
| PII scanning (17 categories) | Pre-call detection with verify + policy actions | 4–8 eng-weeks, FP tuning never ends |
| Governance + audit trails | Per-request lineage, access logs, retention | 2–4 eng-weeks + storage costs |
| Budgets + rate limits + kill-switch | Per-tenant, per-feature hard caps | 2–3 eng-weeks + ops dashboard |
| Tenant isolation + RBAC | Multi-tenant data boundaries | 3–6 eng-weeks, security review required |
| Billing (Stripe + credit ledger) | Per-call charging, monthly grants, overage | 4–8 eng-weeks + reconciliation logic |
| PromptOps (registry + versioning) | Review, canary, audit for prompts | 3–6 eng-weeks |
| Operational support + SLA | On-call, incident response, vendor relations | Ongoing, 1 FTE implicit |
Summed: roughly 5–9 engineer-months to reach feature parity, plus ongoing ops cost. During the design-partner phase you get these layers bundled without a per-call markup.
Total cost of ownership: build vs SchneeAI
The interactive calculator below lets you model the trade-off in real time. Set team size, expected monthly traffic, and the layers you would need to build yourself; the totals update as you change them. Scenario buttons give you starting points.
The model behind it is conservative: eng-weeks come from the build-it-yourself column above, monthly maintenance is a per-layer FTE fraction, and the parallelism cap reflects that beyond ~4 engineers, coordination overhead offsets throughput. SchneeAI’s monthly cost during the design-partner phase equals provider fees (1× passthrough). Both sides pay the same provider; the difference is the build overhead — eng-weeks, maintenance, on-call, infrastructure.
When SchneeAI is the wrong answer
We will say it directly:
- Single-tenant internal tool with no compliance needs — use LiteLLM directly
- Sub-scale POC — use raw provider APIs until you know what you need
- Already built the platform — if you have Vault, billing, PII scanning, and PromptOps running, SchneeAI would duplicate them
SchneeAI is worth it when you would otherwise build (or have already built but struggle to maintain) the layers above.
How to read the calculator numbers
- Per request — cost for one call at the given token counts (provider list price during design-partner phase).
- Input / output rate — provider list rate per 1M tokens.
- Per 1K requests — useful for comparing cost-per-call between models.
- Passthrough badge — indicates 1× passthrough during the design-partner phase; the platform fee is under adjustment.
- Monthly estimate — only shown when you set daily traffic above zero.
Pricing notes
These numbers move. Treat the calculator as an estimate, not a quote. Specifically:
- Provider list prices change — DeepSeek has revised rates multiple times within a single quarter, and new model launches (Gemini, Claude, GPT) typically arrive with promotional pricing that gets adjusted later. Anthropic and OpenAI have both raised rates on legacy models and cut rates on new ones inside the same calendar year.
- FX rates fluctuate daily — the JPY/USD rate has swung between ¥140 and ¥160 over the past year. The calculator uses the value you set in the FX field (defaults to ¥150).
- Platform fee is under adjustment — during the design-partner phase SchneeAI passes provider cost through at 1× (no markup). The platform fee will be set as the product matures; existing design partners are price-locked for their first year.
- Your actual billed credits depend on workload shape — cache-friendly prompts with cheap default models cost less than the worst case; cache-hostile workloads with strong models cost more. Prompt caching, batch discounts, and volume commits are not modelled here, but apply on top with or without SchneeAI.
- Long-context tiering — Gemini 2.5 Pro (2M tokens) and GPT-4o (128K) have tiered rates above 200K tokens. This calculator uses the standard tier for simplicity.
The _meta.updated field in pricing.json shows when we last synced. We aim to refresh within a week of any provider announcing a rate change. For final billed credits, see the pricing page or your service’s usage dashboard.
Use cases
- Model selection — compare the cost of equivalent workloads between Flash, Sonnet, and DeepSeek. For the framework on when reasoning models earn their cost, see Reasoning models: when to pay.
- Budgeting — project monthly spend before onboarding a new feature.
- Build-vs-buy sanity check — compare the calculator’s monthly estimate against what your team would spend building the platform in-house.