MMOOZZGG.com

Comparison

LLM API price comparison in US dollars: 17 models on one date

3 min read13 sources

Providers publish prices in different units: dollars per million, Russian currency per thousand, packages. For this LLM API price comparison, everything is shown in US dollars per million tokens: prices as of 29 September 2026, converted at the Bank of Russia rate of 82.98 per dollar on 13 November 2026 and rounded, with input and output shown separately and a date and source for each line. This page compares facts and does not recommend a model.

On this page · 5 sections

Flagships: $12–$51 per million output tokens

Claude Fable 5.1: $10 input, $51 output. GPT-6 Astra: the same $10 and $51. GPT-5.5: $5 and $31. Claude Opus 5.5: $4 and $20. Gemini 3.1 Pro: $2 and $12, and with context over 200,000 tokens $4 and $18. Flagships are chosen for complex reasoning, long multi-step tasks and code; for a support chat they are more than you need.

Mid-range: $4–$10 per million output tokens

Claude Sonnet 5.5 and GPT-6 Sol: $2 input, $10 output. Claude Haiku 4.5: $1 and $5. GPT-5.4 mini: about $0.75 and $5. Gemini 3.8 Flash: about $0.75 and $4 on a promotion until 31 December 2026. This class is often chosen for working agents: data extraction, classification, answers from a knowledge base.

Cheap: up to $2 per million

GPT-6 Luna: about $0.10 input, about $0.50 output. Gemini 3.1 Flash-Lite: about $0.25 and $2. The Lite model of a large Russian bank: about $0.80 per million for input and output. At these prices, 60 million tokens a month cost $12–$47, and by our calculation local hardware does not pay back within 36 months; these models are chosen for mass simple tasks.

Russian models: about $0.80–$14 per million

The Pro model of a major Russian search company: $14 per million synchronous and $7 asynchronous, input and output at one price; its Lite version $2 and $1. A large Russian bank’s models: Max $8, Pro $6, Lite about $0.80 synchronous, half price asynchronous; packages: Pro 50 million tokens for $300, that is $6 per million, Max 30 million for $250, Lite 300 million for $250, with a minimum payment of $7 a month; in February 2026 the bank cut its prices threefold. Russian mid-range models cost as much for input as foreign flagships or more, but they are paid in Russian currency under a contract with a Russian legal entity, and the data stays in Russia.

What stands behind the price

Payment for foreign APIs from Russia goes through intermediaries who charge a fee, Russian addresses are not on the lists of supported regions, and OpenAI has blocked API traffic from unsupported countries since 9 July 2024. According to the analysis that Anti-Malware retells, most foreign model vendors offer no data processing agreement under 152-FZ, Russia’s personal data law. Prices change: Gemini 3.8 Flash is on a promotion until the end of the year, the Russian bank cut its prices threefold in February, and the Bank of Russia rate changes every day. The date on this page is its expiry date.

Questions and answers

Which model is the cheapest?

By price per token as of 29 September 2026: GPT-6 Luna, then Gemini 3.1 Flash-Lite and the Lite model of a large Russian bank. But a cheap model that makes mistakes costs more: count the cost of a solved task, and a pilot shows it.

Why do Russian models charge the same for input and output?

That is how their billing works: one price per 1,000 tokens. Foreign models charge 4.5–6 times more for output than for input, so for long documents with short answers foreign models are relatively cheaper.

Sources and fact-checking

Checked by the editors on 29 September 2026.

  1. Bank of Russia, exchange rates as of 29 September 2026 (cbr.ru)
  2. Claude Platform, API prices (checked 29 September 2026) (platform.claude.com)
  3. Puter, review of OpenAI API prices based on openai.com, 29 September 2026 (developer.puter.com)
  4. Puter, review of Gemini API prices, 8 September 2026 (developer.puter.com)
  5. kalinkindev.ru, API prices of a Russian search company’s model from its cloud billing, 13 September 2026 (kalinkindev.ru)
  6. A large Russian bank’s developer portal, API rates of its model for companies, 16 September 2026 (developers.sber.ru)
  7. Habr, a large Russian bank cut its model API prices threefold, 4 February 2026 (habr.com)
  8. Claude Platform, supported regions (checked 29 September 2026) (platform.claude.com)
  9. OpenAI, supported API countries (checked 29 September 2026) (developers.openai.com)
  10. Caixin, OpenAI blocks API traffic from unsupported countries from 9 July 2024, 26 June 2024 (caixinglobal.com)
  11. Anti-Malware, 152-FZ and neural networks, 17 June 2026 (anti-malware.ru)
  12. Bank of Russia, official dollar rate as of 29 September 2026 (cbr.ru)
  13. A Russian search company’s AI platform, rates, 30 September 2026 (aistudio.yandex.ru)

Found a mistake? Tell us

Publications

Contact us