ИОАННIOANN
RUEN
Menu

LLM API cost calculator

Estimate monthly model costs using GigaChat API and Yandex AI Studio prices. Enter your workload and response length to compare models. Self-hosted server costs are covered separately below.

Your workload

Adjust the values. The estimate updates automatically.

user requests or events

Instructions, conversation history and retrieved text

Length of the model's answer

Retries, checks and agent steps

30, or 22 for working days

A token is a piece of text. Use token counts from real requests for a closer estimate.

Monthly estimate

33,000 model calls, including retries, checks and agent steps

Synchronous API prices, VAT included. Model calls only.

  • GigaChat Lite

    GigaChat

    Per month
    4,076 ₽
    Per 1,000 calls
    124 ₽

    Per 1K tokens, input / output: 0.065 / 0.065 ₽

  • Alice AI LLM Flash

    Yandex AI Studio

    Per month
    7,590 ₽
    Per 1,000 calls
    230 ₽

    Per 1K tokens, input / output: 0.1 / 0.2 ₽

  • YandexGPT Lite

    Yandex AI Studio

    Per month
    12,540 ₽
    Per 1,000 calls
    380 ₽

    Per 1K tokens, input / output: 0.2 / 0.2 ₽

  • Qwen3.6 35B

    Yandex AI Studio (open model hosted in AI Studio)

    Per month
    13,860 ₽
    Per 1,000 calls
    420 ₽

    Per 1K tokens, input / output: 0.2 / 0.3 ₽

  • gpt-oss-120b

    Yandex AI Studio (open model hosted in AI Studio)

    Per month
    18,810 ₽
    Per 1,000 calls
    570 ₽

    Per 1K tokens, input / output: 0.3 / 0.3 ₽

  • GigaChat Pro

    GigaChat

    Per month
    31,350 ₽
    Per 1,000 calls
    950 ₽

    Per 1K tokens, input / output: 0.5 / 0.5 ₽

  • Alice AI LLM

    Yandex AI Studio

    Per month
    40,590 ₽
    Per 1,000 calls
    1,230 ₽

    Per 1K tokens, input / output: 0.5 / 1.2 ₽

  • GigaChat Max

    GigaChat

    Per month
    40,755 ₽
    Per 1,000 calls
    1,235 ₽

    Per 1K tokens, input / output: 0.65 / 0.65 ₽

  • YandexGPT Pro 5.1

    Yandex AI Studio

    Per month
    50,160 ₽
    Per 1,000 calls
    1,520 ₽

    Per 1K tokens, input / output: 0.8 / 0.8 ₽

An estimate of model call costs. Minimum charges and other running costs are explained below. Prices checked on 9 October 2026. Sources: GigaChat API: тарифы для юридических лиц, Yandex AI Studio: правила тарификации.

How to read the result

  • Each provider counts tokens with its own tokenizer, so the same text yields a different token count. The calculator uses one count for all, so compare orders of magnitude, not exact figures.
  • GigaChat API for companies has a minimum charge of 600 ₽ per month incl. VAT in months with usage. Asynchronous mode is roughly half the price at both providers when an immediate answer is not required.
  • This covers model calls only. A running service also pays for embeddings and retrieval for RAG, storage, monitoring and the team's maintenance time.

When to consider a self-hosted model

A model on your own infrastructure is considered for two reasons: data cannot leave your perimeter, or the load is steadily high and the API bill grows faster than hardware cost would. It cannot be computed from a template: a few factors decide it.

  • Model size and the quality you need: they determine the GPUs and how many.
  • Utilisation: a server costs the same day and night, while an API bills only requests.
  • Peaks: bursts need spare capacity or a queue.
  • Operations: model updates, monitoring, redundancy and on-call.
Review your scenario with usWhat an AI agent costs

FAQ

How many tokens are in one request?

Characters per token depend on the model and the language, so there is no universal ratio. It is more reliable to send 20 to 30 real requests from your scenario through the API and read the token counts the service returns.

Why are OpenAI and other foreign models not included?

The calculator targets companies operating in Russia, where paying for foreign APIs needs a separate arrangement. Foreign platforms are compared in our AI agent platform overview.

What are retries and why count them?

These are calls the system sends again after a timeout, an error or an answer check. In agents one user request often means several model calls. Leave them out and the first invoice will be higher than planned.

Comparing foreign platforms? AI agent platforms in 2026 →

Have a scenario and a budget question? We'll estimate the full running cost, not just tokens.

Start a project