Skip to main content
Fhddos lets you export a detailed CSV log of every API request your account makes. Each row contains the raw token counts and unit prices needed to reproduce your invoice line by line. Use the field definitions and per-model formulas on this page to audit any charge and confirm it matches what Fhddos billed.

​
Export Field Descriptions

After downloading a CSV export from Console → Logs → Export, you will find the following columns. Use them as inputs to the formulas in the sections below.
FieldDescription
Prompt TokensTotal upstream input tokens returned by the provider — use this for cross-checking, not for billing calculation
Input TokensNon-cached text input billed at the standard text rate; equals Prompt Tokens when there is no caching and no image input
Cache Hit TokensInput tokens served from cache — billed at the cache-read discount rate
Cache Write TokensTokens written to a short-lived cache (5-minute TTL) — billed at the cache-write rate
Cache Write 1h TokensTokens written to an extended cache (1-hour TTL) — billed at the 1-hour cache-write rate
Output TokensTotal output tokens including any reasoning tokens produced by the model
Reasoning TokensModel-internal reasoning output included within Output Tokens — billed at the reasoning rate on supported models
Input Unit PriceCNY per million Input Tokens
Output Unit PriceCNY per million Output Tokens (non-reasoning portion)
Cache Hit Unit PriceCNY per million Cache Hit Tokens
Cache Write Unit PriceCNY per million Cache Write Tokens (5-minute)
Cache Write 1h Unit PriceCNY per million Cache Write 1h Tokens
Reasoning Unit PriceCNY per million Reasoning Tokens
Search CountNumber of Claude Web Search tool calls — billed per call
Search Unit PriceCNY per Search call
Image Input TokensImage-modality input tokens (GPT-Image series)
Image Input Unit PriceCNY per million Image Input Tokens
Cached Image TokensCache-hit image input tokens (GPT-Image series)
Cached Image Unit PriceCNY per million Cached Image Tokens

​
Verification Relationship

Use Prompt Tokens to confirm that all input token sub-fields add up correctly before running a billing formula. Plain text requests: 
Prompt Tokens = Input Tokens + Cache Hit Tokens + Cache Write Tokens + Cache Write 1h Tokens
Requests with image input: 
Prompt Tokens = Input Tokens + Cache Hit Tokens + Cache Write Tokens + Cache Write 1h Tokens
              + Image Input Tokens + Cached Image Tokens
Prompt Tokens are for verification only. Always compute amounts by multiplying each sub-field by its corresponding unit price — do not multiply Prompt Tokens directly by any unit price.

​
Precision Note

Formula amounts may differ from your actual invoice by less than ¥0.0001. This small discrepancy comes from internal quota ceiling rounding (ceil) and is expected behavior, not a billing error.

​
Claude Series

Applies to: claude-sonnet-4-6, claude-opus-4-6, claude-haiku-4-5, and similar models. Claude caching splits into two TTL tiers (5-minute and 1-hour). Cache writes cost more than regular input, but cache reads provide a substantial discount. Web Search is an optional add-on billed per call.

​
Billing Formula

Amount = Input Tokens/1M × Input Unit Price
       + Cache Hit Tokens/1M × Cache Hit Unit Price
       + Cache Write Tokens/1M × Cache Write Unit Price
       + Cache Write 1h Tokens/1M × Cache Write 1h Unit Price
       + Output Tokens/1M × Output Unit Price
       + Search Count × Search Unit Price
Search Count and Search Unit Price appear in the export only when the request used the Claude Web Search tool. Leave them as 0 if the columns are empty.

​
Worked Example — claude-haiku-4-5

FieldValue
Prompt Tokens58,518
Input Tokens465
Cache Hit Tokens11,944
Cache Write Tokens29,762
Cache Write 1h Tokens16,347
Output Tokens572
Input Unit Price¥3.5 / M
Cache Hit Unit Price¥0.35 / M
Cache Write Unit Price¥4.375 / M
Cache Write 1h Unit Price¥7.0 / M
Output Unit Price¥17.5 / M
465 / 1M × 3.5       = ¥0.001628
11,944 / 1M × 0.35   = ¥0.004180
29,762 / 1M × 4.375  = ¥0.130209
16,347 / 1M × 7.0    = ¥0.114429
572 / 1M × 17.5      = ¥0.010010
─────────────────────────────────
Total                  = ¥0.260456

​
GPT-5 Series

Applies to: gpt-5, gpt-5.4, gpt-5.4-mini, and similar models. GPT-5 models produce Reasoning Tokens as a sub-component of Output Tokens. Reasoning Tokens are typically priced the same as Output Tokens, but the formula separates them explicitly to handle any future pricing differences.

​
Billing Formula

Amount = Input Tokens/1M × Input Unit Price
       + Cache Hit Tokens/1M × Cache Hit Unit Price
       + (Output Tokens - Reasoning Tokens)/1M × Output Unit Price
       + Reasoning Tokens/1M × Reasoning Unit Price
When Reasoning Unit Price equals Output Unit Price (the default for most GPT models), the formula simplifies to: Input Tokens/1M × Input Unit Price + Output Tokens/1M × Output Unit Price.

​
Worked Example — gpt-5.4

FieldValue
Input Tokens26,672
Cache Hit Tokens0
Output Tokens3,791
Reasoning Tokens2,478
Input Unit Price¥8.75 / M
Output Unit Price¥52.5 / M
Reasoning Unit Price¥52.5 / M
26,672 / 1M × 8.75              = ¥0.233380
(3,791 - 2,478) / 1M × 52.5    = ¥0.068933
2,478 / 1M × 52.5               = ¥0.130095
─────────────────────────────────
Total                             = ¥0.432408

​
GPT-Image Series

Applies to: gpt-image-2, gpt-image-1, and similar image-generation models. Text and image inputs are priced separately. Cached image tokens have their own discount rate. Output Tokens represent image output and use the standard Output Unit Price.

​
Billing Formula

Amount = Input Tokens/1M × Input Unit Price
       + Image Input Tokens/1M × Image Input Unit Price
       + Cache Hit Tokens/1M × Cache Hit Unit Price
       + Cached Image Tokens/1M × Cached Image Unit Price
       + Output Tokens/1M × Output Unit Price
Image billing fields (Image Input Tokens, Image Input Unit Price, Cached Image Tokens, Cached Image Unit Price) are optional export columns. Select them explicitly when configuring your export. These columns are empty for non-image models.

​
Worked Example — gpt-image-2

FieldValue
Input Tokens0
Image Input Tokens10
Cache Hit Tokens10
Cached Image Tokens990
Output Tokens500
Input Unit Price$5.00 / M
Image Input Unit Price$8.00 / M
Cache Hit Unit Price$1.25 / M
Cached Image Unit Price$2.00 / M
Output Unit Price$30.00 / M
0 / 1M × 5.00        = $0.000000
10 / 1M × 8.00       = $0.000080
10 / 1M × 1.25       = $0.000013
990 / 1M × 2.00      = $0.001980
500 / 1M × 30.00     = $0.015000
─────────────────────────────────
Total                  = $0.017073

​
Gemini Series

Applies to: gemini-2.5-flash, gemini-2.5-pro, gemini-2.0-flash, and similar models, including image-generation variants such as gemini-imagen-3. Gemini models support cache reads and cache writes. Some reasoning-capable Gemini models also produce Reasoning Tokens as a sub-component of Output Tokens. Image generation models include multimodal token sub-fields, but these are billed at the aggregate level using the same formula.

​
Billing Formula — Text and Reasoning Models

Amount = Input Tokens/1M × Input Unit Price
       + Cache Hit Tokens/1M × Cache Hit Unit Price
       + Cache Write Tokens/1M × Cache Write Unit Price
       + (Output Tokens - Reasoning Tokens)/1M × Output Unit Price
       + Reasoning Tokens/1M × Reasoning Unit Price

​
Billing Formula — Image Generation Models

Image generation models bill on total token counts using the same structure. When Reasoning Tokens is zero (as it is for image-only outputs), the formula simplifies to the plain input + cache + output structure: 
Amount = Input Tokens/1M × Input Unit Price
       + Cache Hit Tokens/1M × Cache Hit Unit Price
       + Cache Write Tokens/1M × Cache Write Unit Price
       + Output Tokens/1M × Output Unit Price
When a Gemini image request is blocked upstream or produces no image but the provider still returns prompt token usage, Fhddos records and bills those input tokens. The API response will include a usage object for alignment with your consumption logs. Only requests where Fhddos has settled a charge will carry this usage payload — unauthenticated failures, local validation failures, or requests where the upstream returned no usage data are not billed and carry no usage field.

​
DeepSeek Series

Applies to: deepseek-v4-pro, deepseek-v4-flash, and similar models. DeepSeek supports cache reads with a significant discount. There are no write-cache tiers, and no reasoning or image sub-fields.

​
Billing Formula

Amount = Input Tokens/1M × Input Unit Price
       + Cache Hit Tokens/1M × Cache Hit Unit Price
       + Output Tokens/1M × Output Unit Price

​
Universal Billing Template

Apply this single formula to any model in the Fhddos catalog. Treat any empty export field as 0 — it will not affect the result. 
Amount = Input Tokens/1M × Input Unit Price
       + Image Input Tokens/1M × Image Input Unit Price
       + Cache Hit Tokens/1M × Cache Hit Unit Price
       + Cached Image Tokens/1M × Cached Image Unit Price
       + Cache Write Tokens/1M × Cache Write Unit Price
       + Cache Write 1h Tokens/1M × Cache Write 1h Unit Price
       + (Output Tokens - Reasoning Tokens)/1M × Output Unit Price
       + Reasoning Tokens/1M × Reasoning Unit Price
       + Search Count × Search Unit Price
Paste this formula directly into a spreadsheet and reference each CSV column. Set any empty price column to 0. Image input fields carry values only for the GPT-Image series; Search fields carry values only for Claude requests that invoke Web Search.