APIs and development

76 AI Responses Cost About 6 Rubles: What AI Use Really Costs

In a small sample of 76 generations, the API reported a total cost of $0.0716—about RUB 6 at an exchange rate of RUB 85 per dollar. This is not a universal price for 76 requests: the total depends on the model, context, answer length, and reasoning tokens.

76 generations cost about RUB 6

The sample included 76 generations for which the provider API returned cost information. The total was $0.0716, or about RUB 6 at RUB 85 per dollar.

The sample is small and does not represent every set of requests. It does give a sense of scale: in these examples, an individual generation cost from a few kopecks to a few rubles.

What individual answers cost

The GPT-6 Luna requests were not all very short: one generation added 770 input tokens, while another returned 549 output tokens.

One Gemini 3.8 Flash request used 11 input and 212 output tokens, including 195 reasoning tokens. It cost about seven kopecks.

Four Grok 4.7 generations cost about 48 kopecks in total. Their contexts were larger, with roughly 1,200 new input tokens per request.

Claude Opus 5.5 was the most expensive example. One answer used 55 input tokens, 2,254 output tokens, and 350 reasoning tokens. The API estimated the generation at $0.0453, or about RUB 3.90. The model was used to write a small Python snippet.

ModelUsageCost
GPT-6 Luna12 requestsabout RUB 0.11
Gemini 3.8 Flash1 requestabout RUB 0.07
Grok 4.74 requestsabout RUB 0.48
Claude Opus 5.51 long answerabout RUB 3.90

Why costs vary so much

A token is a small piece of text, not always a whole word. Models charge different rates for input and output tokens, and the total also depends on how much text the model generates.

That is why a message does not have one fixed price. A short question with a short answer may cost a few kopecks. A large context, long generation, and model reasoning can multiply usage tens or even hundreds of times.

Reasoning tokens are part of the computation a model uses for internal reasoning. Some APIs report them separately even though the user sees only the final answer.

Does that mean a $20 subscription is a bad deal?

Not necessarily. A fixed subscription may be better value if you code constantly, send large documents, use long context, and make hundreds of requests.

A typical user may have a different pattern: a few questions each day, editing, translation, research, or help with a spreadsheet or code. Compare your actual monthly usage rather than an abstract subscription price against API pricing.

A small sample cannot support a final conclusion. But API use can be very inexpensive for lightweight requests.

The cheapest model is not always the best value

Generation cost alone says little about how useful an answer is. If a cheaper model needs three follow-up prompts while a more expensive one solves the task on the first try, the overall economics change.

A more useful question is: how much does it cost to get an answer that actually solves the task? Send the same real requests to several models and compare both token costs and answer quality. For example, ten tasks across ten models give you 100 answers to assess under the same conditions.

How to compare models on your own requests

Pick a few tasks you do regularly, then compare the answers and actual spending. The Any LLM model catalog shows the current model list, and you can submit the same request in the chat. Availability and terms can change.

For context, see our reviews of GPT, Claude, Gemini, and Grok. Token pricing is only one part of the comparison.

Where the data came from

We tracked spending through Any LLM, where several AI models are available in one chat. The calculations use the costs returned by provider APIs for those specific generations.

This is not a reconciliation against a provider’s final invoice or a universal rate for each model. The figures describe only these requests. Check the current prices and terms of the specific API when planning a budget.

Sources

How we check our guides