AI Relay Charges, Tokens and Ten-Request Reconciliation
Review ten-request account ledgers and meters across six paid AI relay test accounts, matching token and cache counts, same-route calculations, observed charges, and clearly labeled inferred rules.
Account usage is not wallet balance
Some provider-compatible endpoints return cumulative usage. The number rises as charges accrue; this report preserves the original meter type instead of calling it remaining balance.
One reconciled charge does not verify every route
A ten-request ledger applies only to the recorded key, model, protocol, and group. Matching the same-route calculation does not establish another route; inferred cache rates remain labeled as inference.
Reconciliation coverage
- Paid test providers
- 6
- Completed ten-request sessions
- 7
- Readable cumulative-usage sessions
- 4
- Latest measured session
- 2026-09-19
The before/after values are a provider account's cumulative usage meter, not its remaining wallet balance. The difference matches the recorded session charge. A short session does not establish a per-million-token rate.
Observed ten-request charges
| Provider | Model / protocol | Date | Passed | Usage before | Usage after | Meter difference | Input / output tokens | Cache-read tokens | Published-price reconciliation |
|---|---|---|---|---|---|---|---|---|---|
| APINebula # | gpt-6-astraOpenAI | 2026-09-19 | 10 / 10 | ¥0.296454 | ¥0.302154 | ¥0.0057 | 1,368 / 107 | 0 | Account ledger reviewed below |
| APINebula # | claude-fable-5-1ClaudeCode | 2026-09-19 | 10 / 10 | ¥0.31723 | ¥0.36599 | ¥0.04876 | 2,488 / 76 | 0 | Public meter and historical quote matched |
| LinksAPI # | gpt-6-astraResponses | 2026-09-18 | 10 / 10 | $0.214796 | $0.27286 | $0.058064 | 42,732 / 79 | 39,808 | Account ledger reviewed below |
| boxying # | gpt-6-astraOpenAI | 2026-09-18 | 10 / 10 | ¥0.118548 | ¥0.141152 | ¥0.022604 | 42,508 / 70 | 40,064 | Account ledger reviewed below |
Input tokens may include cache-read tokens; do not add the two counts to estimate a bill. The separate tables below distinguish public cumulative meters from account-ledger reviews and state each remaining limitation.
Reviewed same-route price snapshot
| Provider | Model / account group | Request date | Input / output tokens | Account log / public meter | Earlier quoted rate calculation | Quote / log review dates | Quoted input / output per 1M |
|---|---|---|---|---|---|---|---|
| APINebula # | claude-fable-5-1Claude Code global high-availability route | 2026-09-19 | 2,488 / 76 | ¥0.04876 | ¥0.048756 | 2026-09-182026-09-22 | ¥17 / ¥85 |
The reviewed ten-request account-log total matches the public cumulative-usage meter and the model/token totals. No cache-read tokens were recorded. The earlier provider-owned quote for the same named group yields the displayed calculation; the difference is within the account display rounding. Raw private account logs are not published. This does not verify today's rate, wallet balance, model identity or subscription charges.
Reviewed account-ledger sessions
| Provider | Model / tested-key group | Request date | Input / output / cache | Ten-call ledger | Rate calculation | Input / output / cache per 1M | Rate evidence / reviewed | Evidence boundary |
|---|---|---|---|---|---|---|---|---|
| UU API # | gpt-6-astraCodex-GPT Pro pool | 2026-09-27 | 42,508 / 70 / 40,192 | $0.026738 | $0.026741 | $4 / $20 / $0.4 | inferred_historical_account_log2026-09-27T13:26:13.682Z | Ten historical account-log rows show the same key, Responses endpoint, Codex-GPT Pro pool and per-call charges, totaling USD 0.026738. Effective input, output and cache rates are inferred from those rows and are not presented as provider-published rates. |
| H API # | claude-fable-5-1Claude Max external route | 2026-09-18 | 2,928 / 121 / 0 | $0.063594 | $0.063594 | $18 / $90 / $0 | Later current quote matches2026-09-24 | The current same-group quote reviewed after the session exactly matches the ten-call log; it does not prove that the historical rate never changed. |
| OpenOx # | gpt-6-astraGPT unified / limited-time route | 2026-09-19 | 42,508 / 70 / 40,320 | $0.052559 | $0.05256 | $8 / $40 / $0.8 | Historical per-call ledger2026-09-24 | The historical per-call log shows input, output and cache-read rates; the calculation differs from the ten-call charge by USD 0.000001. |
| boxying # | gpt-6-astracodex_pro pool, 0.4x | 2026-09-18 | 42,508 / 70 / 40,064 | ¥0.027206 | ¥0.027202 | ¥4 / ¥20 / ¥0.4 | Historical label and multiplier2026-09-24 | The historical log shows the standard rate, tested-key group and 0.4x multiplier; the calculation differs only by displayed rounding. |
| LinksAPI # | gpt-6-astraGPT Pro (Codex included), 0.3 reference / 0.5x billing | 2026-09-18 | 42,732 / 79 / 39,808 | $0.058064 | $0.058064 | $6.25 / $31.25 / $0.9375 | Cache rate inferred2026-09-24 | The log shows the tested-key group, standard input/output rates and 0.5x billing multiplier. The cache-read rate is inferred from the ten-call ledger, not presented as an explicit provider quote. |
| APINebula # | gpt-6-astraCODEX, 0.37x | 2026-09-19 | 1,368 / 107 / 0 | ¥0.007044 | ¥0.007041 | ¥3.7 / ¥18.5 / ¥0 | Historical per-call ledger2026-09-24 | The historical per-call log shows the base rates and 0.37x group multiplier; the small aggregate difference comes from per-call rounding. |
| APINebula # | claude-fable-5-1Claude Code global high-availability route, 1.7x | 2026-09-19 | 2,488 / 76 / 0 | ¥0.04876 | ¥0.048756 | ¥17 / ¥85 / ¥0 | Historical per-call ledger2026-09-24 | The historical per-call log shows the base rates and 1.7x group multiplier; the ten-call calculation matches the account charge. |
These rows were matched to the public ten-request sessions by provider, model, timestamp and token totals. Raw private ledgers, account identifiers, IP addresses and credentials remain private. A later current quote is not treated as a historical snapshot, and an inferred cache rate is labeled as inference.
Completed calls without a readable meter
| Provider | Model | Date | Evidence state |
|---|---|---|---|
| UU API | gpt-6-astra | 2026-09-27 | 10 requests completed; account meter unavailable |
| OpenOx | gpt-6-astra | 2026-09-19 | 10 requests completed; account meter unavailable |
| H API | claude-fable-5-1 | 2026-09-18 | 10 requests completed; account meter unavailable |
A successful session without a readable meter has no observed charge. Neither a list price nor a full-use subscription estimate fills that gap.
Source, method and limits
Each session uses an AI Relay Check test account and ten requests. The source records model ID, protocol, outcomes, tokens, account-meter readings and timestamp. Open a provider profile for public request-level evidence. Test credit may include recharges and promotional amounts.
To complete theoretical reconciliation, record the exact route group, its price source and capture date, input/output/cache rules, and matching meter or wallet readings. Subscription equivalents additionally require actual eligible quota consumption.
The tables on this page retain privacy-reviewed session totals, token fields, account-meter differences and reconciliation limits. Request-level records can be checked through each provider profile and its public evidence endpoint. Private account logs and credentials are not published.
Download privacy-reviewed billing reconciliation (CSV) · Dated evidence changes (CSV) · Provider summary (JSON)
Partner links and test credit, where present, are disclosed on provider profiles. Corrections: airelaycheck@163.com
Frequently asked questions
Is the before/after value a remaining wallet balance?
No. The measured sessions use a cumulative account-usage meter, which increases as charges accrue. Remaining wallet balance needs separate evidence.
Does the session charge verify the advertised token rate?
Several ten-call account ledgers now agree with their same-route rate evidence within displayed rounding. Each row states whether the rate came from the historical ledger, a later current snapshot, or an inferred cache rule. None verifies other routes or subscription billing.
Why do two providers show completed sessions without a charge?
Ten requests completed, but their account meter was not readable. Request success does not establish a billed amount.
Answer summary
- Latest test date
- Providers covered
- 6
- Models / protocols
- claude-fable-5-1 / gpt-6-astra · ClaudeCode / OpenAI / Responses
- Successful / total samples
- 7 / 7
- Price / billing verified
- 6/6 price · 6/6 billing
- Still unverified
- Current rates, other routes, model identity and subscription billing
- One-sentence conclusion
- Use only the rows with matching route and ledger evidence as billing verification; completed requests without a readable meter remain unverified.