GigaChat 3.5 Ultra API — price in US dollars

What GigaChat 3.5 Ultra is

GigaChat 3.5 Ultra is the flagship model of Sber’s GigaChat family. It has a 262,144-token context window, takes text in and gives text out, and belongs to the chat and text generation category. Through kosareva.cloud it is available via one OpenAI-compatible API: the same code as for OpenAI — just change base_url and the key.

GigaChat is not on OpenRouter as of October 2026, and a direct contract for the GigaChat API needs a Russian company or sole proprietor. Here a company outside Russia gets GigaChat 3.5 Ultra with an English invoice in US dollars from Kosareva Cloud LLC (Armenia) and pays by bank transfer. Requests to GigaChat 3.5 Ultra are processed in Russia.

One key opens more than GigaChat 3.5 Ultra: the catalog has over a hundred models, including the other GigaChat and Yandex models. That matters more than it seems: models age within months, and moving to the next version is one line of code, not a new contract and a new integration.

GigaChat 3.5 Ultra strengths and weaknesses

Who it suits. Long documents and correspondence in Russian, answers from an internal knowledge base, scenarios with tool calls. And any task where the customer requires the data to stay in Russia.

What to compare it with. Against YandexGPT 5.1 Pro ($13.10 / $13.10): eight times cheaper on input with an eight times larger context. Against Alice AI ($8.19 / $19.65): five times cheaper on input and four times on output, with twice the context. All three are processed in Russia, so the choice here is about price and context length, not compliance.

GigaChat 3.5 Ultra among the GigaChat models

GigaChat 3.5 Ultra is the flagship of the family. GigaChat 3 Pro has the same 262K context and is a quarter cheaper per million input tokens ($1.20 against $1.58), but its service prompt is about 1,214 tokens against Ultra’s 65 — so on requests shorter than about 3,500 tokens Ultra comes out cheaper. GigaChat 3 Lightning costs $0.20 per million but is a small model with a 990-token service prompt; GigaChat 2 Max is six times more expensive on input with half the context.

GigaChat 3.5 Ultra price in US dollars

Approximate: the catalog price (134.71 ₽ input and 404.04 ₽ output per 1M tokens) divided by 85.5 ₽ per $1, ID Bank’s non-cash rate on 3 October 2026. Your balance is kept in rubles; a USD invoice is converted at ID Bank’s non-cash rate on the invoice date. The vendor’s service prompt, about 65 tokens per request, is billed as input. You pay only for actual usage.

What a typical task costs on GigaChat 3.5 Ultra

A price per million tokens says little on its own, so here are four common tasks at GigaChat 3.5 Ultra rates: $1.58 per million input tokens and $4.73 per million output tokens, plus the vendor’s service prompt of about 65 input tokens on every request.

Calculated from the current kosareva.cloud price list, in US dollars at 85.5 ₽ per $1, with the service prompt included.

GigaChat 3.5 Ultra cost calculator

Move the sliders to your volume — the monthly cost in US dollars updates at once, with no sign-up.

At $1.58 per million input tokens and $4.73 per million output tokens; approximate, at 85.5 ₽ per $1. Each request also carries the vendor’s service prompt of about 65 input tokens — it is already in the numbers. Billed for actual usage.

How to connect GigaChat 3.5 Ultra

Use the standard OpenAI SDK with the kosareva.cloud base_url and the model gigachat-3-5-ultra. In existing code you change exactly two lines: the address and the key.

Python, OpenAI SDK

from openai import OpenAI

client = OpenAI(
    api_key="YOUR_KOSAREVA_KEY",
    base_url="https://api.kosareva.cloud/v1",
)

response = client.chat.completions.create(
    model="gigachat-3-5-ultra",
    messages=[{"role": "user", "content": "Hello!"}],
)
print(response.choices[0].message.content)

Python without the SDK: requests

import requests

response = requests.post(
    "https://api.kosareva.cloud/v1/chat/completions",
    headers={"Authorization": "Bearer YOUR_KOSAREVA_KEY"},
    json={
        "model": "gigachat-3-5-ultra",
        "messages": [{"role": "user", "content": "Hello!"}],
    },
    timeout=600,
)
print(response.json()["choices"][0]["message"]["content"])

cURL

curl https://api.kosareva.cloud/v1/chat/completions \
  -H "Authorization: Bearer YOUR_KOSAREVA_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"gigachat-3-5-ultra","messages":[{"role":"user","content":"Hello!"}]}'

The same key works with any client that supports an OpenAI-compatible endpoint: LangChain, LlamaIndex, Cursor, Continue, n8n. You don’t need a separate key per model — one key opens the whole catalog.

GigaChat 3.5 Ultra endpoints and parameters

The gateway follows the OpenAI schema, so parameter names and the response format are the ones you know. One base address for every model in the catalog.

Request long answers with stream: true: with a large max_tokens, a regular request stays silent until generation ends and can hit your HTTP client’s timeout. In streaming mode the first tokens arrive within seconds, and the timeout problem goes away.

Specifications

What GigaChat 3.5 Ultra is good for

• Chatbots, assistants, Q&A and customer support in Russian
• Long documents and correspondence in Russian: summaries, answers from a knowledge base
• Scenarios with tool calls

What to keep in mind with GigaChat 3.5 Ultra

The service overhead on this model is small — about 65 tokens by our measurement, so a short question costs like a short question. On GigaChat 3 Pro the overhead is almost twenty times larger, and on short requests that outweighs the difference in price per million. Before choosing between them by the price list, run your typical request through both and compare the actual bill.

First, context runs out faster than it seems. In a dialogue the whole conversation goes into the input, not just the last message, so a long chat gets more expensive with every turn. If the history isn’t needed for the answer, trim it: savings on input tokens usually matter more than picking a cheaper model.

Second, the 262K-token window is shared between the request and the answer: the more you send, the less room the model has to generate. Set max_tokens on purpose, not “with a margin”.

Third, the balance is shared by all models in the catalog. You don’t have to decide in advance how much money to “put on GigaChat 3.5 Ultra”: the same funds go to any other model if you switch tomorrow. Usage per model is shown in the dashboard on a separate line, with the number of requests and tokens.

GigaChat 3.5 Ultra for companies outside Russia

GigaChat 3.5 Ultra, the flagship of Sber’s GigaChat family: a 262K context, tool calling, requests processed in Russia. Access through kosareva.cloud with an OpenAI-compatible API and a USD invoice for companies.

• Register as “Company or sole proprietor” → “In another country”: country of registration, company registration number, tax ID if you have one.
• In “Top up balance”, “Issue invoice (companies)” gives an English invoice in US dollars from Kosareva Cloud LLC (Armenia): minimum $50, due in 10 days.
• Pay by bank transfer to our USD account at ID Bank. The balance is credited after the bank confirms the payment, usually in 1–3 business days.
• Cards issued outside Russia are not accepted. Invoices are in USD; for EUR, write to us.
• Pay only for actual usage, no subscription. One OpenAI-compatible API for GigaChat 3.5 Ultra and over a hundred other models with one key.

Contract and payment. Your contract and payment are with Kosareva Cloud LLC (Armenia), not with the model vendor. You are responsible for compliance with the rules of your jurisdiction; if in doubt, check with your counsel.

Requests to GigaChat 3.5 Ultra are processed in Russia. If data protection rules apply to you, assess the transfer before you send personal data.

Other GigaChat models

Price per 1M input tokens. Output prices and a monthly estimate for each — on the GigaChat page.

GigaChat 3.5 Ultra questions