# We audited 30 days of our own billing (markdown mirror)

Source: https://aiapi-pro.com/blog/billing-audit-zero-markup-2026

Method: take every paid text call served in 30 days, recompute cost from provider list price and token counts, compare to billing records.

## The audit

| | Value |
|---|---|
| Models audited | 58 (with published list price) |
| Realized cost (billing) | $188.74 |
| List-implied cost (tokens x list price) | $188.74 |
| Deviation | **0.005%** (rounding, not markup) |

Largest single-model deviation: 1.72% (GLM-5.1, from per-call rounding to six decimals). Everything else under 0.5%.

## Where the money goes

| | Tokens | Share of tokens | Share of cost |
|---|---|---|---|
| Input | 556.2M | 98.9% | **95.8%** |
| Output | 6.1M | 1.1% | **4.2%** |

Output tokens are priced 2-5x higher per token, so people assume output dominates. In this traffic it does not: input is 95.8% of the bill, because workloads send ~40K input tokens per call and get 815 output tokens back.

Caveat: this is agent/retrieval-heavy traffic (coding tools, RAG, chat-with-docs). Content-generation workloads will look inverted.

## Three consequences

1. **Optimise the prompt before the output.** If input is ~96% of the bill, trimming retrieved context or capping document size saves far more than shortening answers.
2. **Model choice is an 80x lever.** Realized blended cost per 1M tokens ranged from $0.0806 (deepseek-v4-flash, 5,377 calls) to $6.46 (kimi-k2.7-code-highspeed, 37 calls).
3. **Price per token is a starting point.** A call's cost is input_tokens x input_price + output_tokens x output_price; at 40K input per call, even a cheap model has a meaningful per-call cost.

## Reproduce

```
cost = input_tokens * input_price_per_1m / 1e6
     + output_tokens * output_price_per_1m / 1e6

curl -s https://aiapi-pro.com/china-ai-leaderboard.json | jq '.categories[] | .models[] | {name, price_usd}'
```

## FAQ

**Does NovAI add a markup on model prices?**
No. Auditing 30 days of billing across 58 models with published list prices, the realized cost was $188.74 against a list-implied cost of $188.74 - a 0.005% deviation, which is rounding. The gateway bills at provider list price with a 1.0 multiplier.

**Are input or output tokens the bigger cost?**
In this traffic, input. 556.2M input tokens versus 6.1M output tokens meant input was 95.8% of the bill and output 4.2% - even though output tokens are priced 2-5x higher per token.

**Why is input so dominant?**
Agent and retrieval workloads resend a large prompt (averaging about 40K input tokens per call) and receive a short answer. The pattern is common for coding agents, RAG and chat-with-documents products.

**How much does model choice change the bill?**
By up to 80x. In the same 30 days, the realized blended cost ranged from about $0.08 per 1M tokens on deepseek-v4-flash to $6.46 per 1M on kimi-k2.7-code-highspeed.