This guide uses source checks from September 13, 2026. Provider and gateway prices can change; preserve the checked date with every forecast.
Why This Topic Matters Now
The practical buyer question is now simple: listed rates use default ×1, while API keys created with the VIP billing group use ×0.9. The 10% difference still matters in a production forecast, so record the key's effective group beside the dated model row.
This article is for engineering and finance teams evaluating AIWave in Tier 1 and Tier 2 markets. The verification pattern is: read the live pricing feed, record the billing group selected for the API key, send a small placeholder-key request, store usage fields, and attach the route owner. That is the right level of evidence before a larger workload is approved.
Source Facts Checked Today
AIWave /api/pricing checked on September 13, 2026 returned the dated catalog snapshot and group_ratio values of default 1 and vip 0.9. The feed also returned auto_groups with default and a pricing_version hash. Treat those values as publication-day evidence, not a permanent contract. If a finance worksheet uses a multiplier, store the checked timestamp and the API key billing group used for the request sample.
The same live feed listed multiple enabled groups per model. For example, deepseek-v4-pro, deepseek-v4-flash, glm-5.1, kimi-k3, qwen3.6-flash, and step-3.5-flash-2603 were all enabled for default, vip, and svip in the data returned during this run. Enabled group and effective billing group are related fields, but they are not the same evidence. One says a group may access a route; the other says which multiplier applied.
DeepSeek rows checked through AIWave on September 13, 2026 provide a useful worked example. The live row translated deepseek-v4-pro to $1.914 input, $5.742 output, and $0.063736 cache-hit input per 1M tokens before applying any visible group multiplier. A billing review should preserve base row, token usage, cache usage, group multiplier, and final charge as separate columns so the team can reconcile the number later.
Planning Matrix
A source-dated planning matrix keeps the page useful for engineers and procurement reviewers. It turns a search query into an auditable route decision instead of a loose model preference.
| Evidence | Where it comes from | Decision it supports |
|---|---|---|
| pricing_version | AIWave /api/pricing | Which rate set was reviewed |
| base model row | AIWave pricing feed | Token cost before group math |
| effective group | Account or request evidence | Multiplier selection |
| enabled groups | Model metadata | Route availability |
| usage object | API response or logs | Actual token count |
| cache-hit field | Usage and route ledger | Input split |
| route owner | Internal config | Who approves changes |
Implementation Pattern
The implementation pattern keeps credentials as placeholders, pins the AIWave base URL, records the model, and leaves room for route-specific controls. Production applications should move credentials into environment or secret storage.
from decimal import Decimal
AIWAVE_API_KEY = "YOUR_API_KEY_HERE"
def reconcile_request(row, usage, group_multiplier):
input_cost = Decimal(usage["input_tokens"]) / Decimal(1_000_000) * Decimal(str(row["input_per_1m"]))
output_cost = Decimal(usage["output_tokens"]) / Decimal(1_000_000) * Decimal(str(row["output_per_1m"]))
cache_cost = Decimal(usage.get("cache_hit_tokens", 0)) / Decimal(1_000_000) * Decimal(str(row["cache_hit_per_1m"]))
return round((input_cost + output_cost + cache_cost) * Decimal(str(group_multiplier)), 6)
deepseek_pro = {"input_per_1m": 1.914, "output_per_1m": 5.742, "cache_hit_per_1m": 0.063736}
sample_usage = {"input_tokens": 250_000, "output_tokens": 40_000, "cache_hit_tokens": 100_000}
print(reconcile_request(deepseek_pro, sample_usage, group_multiplier=0.9))
Separate Access From Billing
A model row can list several enabled groups. That tells a developer which billing groups can call the route, but it does not prove which group priced one request. Put enabled groups and effective group in different columns. During a trial, confirm the group selected for the API key in Console. If the team cannot confirm it, mark the forecast as provisional and do not use it for procurement approval.
Store the Pricing Version
The live feed exposes pricing_version, which is a useful audit handle. Put that value into the worksheet beside the checked date. A future rate-card update can then be compared against the exact set used in the original decision. This is especially useful for teams that run a one-week proof of concept and then approve production traffic two weeks later. Without the version, a rate change can look like a math error.
Use a Redacted Probe Request
The first probe should use a small redacted prompt, placeholder credential handling, and a route that is already visible to the account. Record status, model, input tokens, output tokens, cache-hit tokens when available, latency, request timestamp, pricing_version, and effective group. Do not send customer content to prove a billing calculation. A small probe is enough to verify compatibility and reconcile the first line item.
Reconcile Before Forecasting
A cost forecast should be built only after at least one request has been reconciled. Start with the base model row, multiply by input and output tokens, apply cache-hit pricing separately, then apply the group multiplier selected for the API key. This process may feel slower than copying a headline number, but it creates a worksheet that engineering, finance, and support can read together.
Key Billing Changes Need Owners
If a key moves from one billing group to another, record the date, reason, owner, and expected effect on forecasts. Do not bury the change inside a pricing cell. A key-group change affects every request sent with that key, so treat it like a billing configuration event. Route owners should review output caps, retry ceilings, and monthly limits after the change.
Internal Links for Billing Searchers
Readers should have a clear path to Pricing, Models docs, Chat Completions docs, Trust, Predictable Pricing, and the 10M-token DeepSeek ledger. Those links move the reader from group concepts to an actual workload estimate.
Procurement Review
Procurement should not approve a large AIWave workload from a screenshot alone. Ask for the live pricing row, pricing_version, checked date, effective group, route name, usage fields, and a reconciliation formula. Engineering should provide the exact test prompt category without exposing sensitive content. Support should be able to explain which row and group were used if the invoice differs from the forecast.
Final Checklist
Before using AIWave group math in a budget, verify access, verify effective group, store pricing_version, capture one redacted request, reconcile tokens, and assign a route owner. Listed rates use default ×1; select VIP when creating the API key to use ×0.9. With those controls in place, billing-group differences become auditable instead of surprising.