API access pricingPer-million-token rate, calculation base and market corridor
We publish the rate openly, because a hidden price list forces the buyer to guess. There is one rule: the vendor's public list price per 1 million tokens, converted at 84 RUB per USD, multiplied by 3.4. That is the upper part of the corridor set by Russian gateways: we are not the cheapest gateway on the market and we do not intend to be. The coefficient drops with prepayment and never goes below 3.0. Data verified on 10 November 2026.
How a vendor price becomes a ruble rate
Three steps you can verify in any row of the table below.
Start from the vendor's public list price
The cost of 1 million tokens per the vendor's public price list on the verification date: input and output separately. No negotiated base rates.
Multiply by 84 and by the coefficient
84 RUB per USD is a fixed reference for comparing rows. The coefficient of 3.4 pays for access: a non-resident contract, the currency payment, rate limits, per-key accounting, support and closing documents.
Fix it in the invoice at the rate of the payment date
The table is a reference. The invoice converts the amount at the bank rate on the payment date, because the rate moves faster than the price list.
Rate by model, RUB per 1 million tokens
Input / output. Coefficient 3.4 is the base rate without prepayment.
| Model | Vendor list, USD | AIO Keys rate, RUB |
|---|---|---|
| Claude Opus 5.5 frontier | 4 / 20 | 1 142 / 5 712 |
| Claude Sonnet 5.5 frontier | 2 / 10 | 571 / 2 856 |
| Claude Haiku 4.5 frontier | 1 / 5 | 286 / 1 428 |
| GPT-5.4 frontier | 2.5 / 15 | 714 / 4 284 |
| Gemini 3 Flash frontier | 0.5 / 3 | 143 / 857 |
| DeepSeek V4 Pro open weights | 1.25 / 3.75 | 357 / 1 071 |
| GLM-5.3 open weights | 1.11 / 3.89 | 317 / 1 111 |
| Qwen3.7-Plus open weights | 0.28 / 1.11 | 79 / 317 |
Cache reads and batch are calculated at the vendor's rate with the same coefficient — there is no separate markup on them. Models not listed here use the same formula; send us the name.
Where this rate sits against the Russian market
Russian gateways sell not a token but the path to a token: ruble payment, a reachable endpoint, closing documents. That is why their retail price exceeds the official-rate conversion — it is an access premium, not a compute premium.
| Claude Sonnet 5, RUB/1M | Input | Output | Vs official conversion |
|---|---|---|---|
| Central Bank conversion | 168 | 841 | 1.00× |
| AITUNNEL | 400 | 2,000 | 2.38× |
| ProxyAPI | 600 | 3,030 | 3.57–3.60× |
| AIO Keys, coefficient 3.4 | 571 | 2,856 | 3.40× |
Public retail rates of Russian gateways as of September 2026; Central Bank conversion as of 23 September 2026 — 84.0657 RUB per USD. Rows are not subtractable from each other: each price carries different commitments on availability, support and documents.
We are not the cheapest gateway in this corridor and we do not aim to be. The 3.4 rate sits right next to ProxyAPI (3.57) and 43% above AITUNNEL (2.38) — by choice. A gateway that prices at the floor sells access without support: no margin left for an integrator's partner discount, no buffer for exchange-rate movement, nobody to call when something breaks at night. We sell volume that does not get switched off mid-project.
Prepayment lowers the coefficient, never below 3.0
The discount is given for predictable volume, not for negotiation. We do not go below 3.0 even on large contracts.
| Prepayment | Coefficient | Claude Sonnet 5.5, in/out |
|---|---|---|
| none, pay as you go | 3.4 | 571 / 2,856 |
| from 1 million RUB | 3.2 | 538 / 2,688 |
| from 5 million RUB | 3.0 | 504 / 2,520 |
Minimum invoice is 30,000 RUB. Prepayment does not expire: the remainder carries over and the usage window is fixed in the contract.
What the coefficient includes and what it does not
The coefficient is not a markup on a token but the price of the wrapper around it. Scope therefore matters more than the number itself.
- Non-resident contract, invoice and acceptance act; the document set for registering the contract with an authorised bank above 3 million RUB.
- Keys with separate per-minute and daily limits, plus per-key and per-project spend accounting.
- OpenAI-standard compatibility: the model name changes, the application code does not.
- Model substitution if a model degrades or is withdrawn — without rebuilding the integration.
- Taxes on the buyer's side: import VAT on services is withheld by you as the tax agent.
- Integration, prompt engineering and scenario acceptance work — quoted separately.
- Guaranteed latency and priority queueing — a separate service level agreement.
- Storage and processing of personal data: such scenarios need an on-premise deployment, not an API.
What this means on an invoice
Profile: an agent on Claude Sonnet 5.5, 4 million input and 1 million output tokens per month.
| Line | At vendor list | At coefficient 3.4 |
|---|---|---|
| Input, 4 million tokens | 672 RUB | 2,285 RUB |
| Output, 1 million tokens | 841 RUB | 2,856 RUB |
| Total per month | 1,513 RUB | 5,141 RUB |
The difference is the price of being able to pay this volume in rubles at all and close it with documents. At 5 million RUB of prepayment the same profile is calculated at a coefficient of 3.0.
What buyers ask about the price
Is this the final price?
This is the public rate. The final amount is fixed in the invoice at the bank rate on the payment date and does not change after the invoice is issued.
Why a coefficient of 3.4 if the token itself costs less?
You pay for access, not for the token: a non-resident contract, the currency payment, rate limits, per-key accounting, support and closing documents. The vendor sells compute; a gateway sells the ability to buy it.
Russian gateways charge 2.4–3.6 times the conversion. Why 3.4 here?
That is an access premium, and we sit in the upper part of the same corridor. We do not go below 3.0 even on large prepayments: at that rate there is no support, no partner discount and no buffer for exchange-rate movement.
You are more expensive than AITUNNEL. Why not move there?
You can, and some buyers do — most come back within two or three months. The price gap is a gap in commitments: incident handling, model substitution when a model degrades, an invoice and act your accounting can book, a partner discount for your integrator. A gateway priced at the floor carries none of that, and when it leaves the market the access leaves with it. We do not sell the cheapest token — we sell volume that does not get switched off mid-project.
Is 84 RUB per USD fixed?
No. It is a reference for comparing rows in the table, not a settlement rate. The invoice amount is calculated at the bank rate on the payment date.
What documents do you provide?
An invoice and an acceptance act from a company registered in Hong Kong: for your accounting this is a foreign trade contract for the import of services. Above 3 million RUB the contract is registered with an authorised bank — your side files it, we submit the document set.
Are cache reads and batch more expensive?
No. They use the same vendor rate with the same coefficient. There is no "cache markup" — otherwise long agent sessions would become uneconomical exactly where they matter.
Can we start without prepayment?
Yes. A pilot volume up to 100,000 RUB is calculated at a coefficient of 3.4 without prepayment. After that — pay as you go, or prepayment with a lower coefficient.