LLM API pricing comparison and cost calculator

How to use it

1. Enter the input and output tokens of one request, or pick a common scenario. If you do not know how many tokens your text is, count it with the token counter on this site.
2. Choose whether to compare in Chinese yuan or US dollars. US providers price in dollars and Chinese providers in yuan; the tool converts with the rate you enter.
3. DeepSeek has peak and off-peak prices. By default the tool uses the current Beijing time to decide which applies; you can also set it by hand.
4. The table is sorted by the cost of this request, cheapest first. "List price" shows each provider's original price and currency; "Context used" says "Does not fit" if the request exceeds the context window or output limit.

Where the prices come from

Models and sources

Chinese providers: DeepSeek V4 Pro and V4.1 Flash; Qwen3.8 Max and Qwen3.8 Flash on Alibaba Cloud Model Studio (Beijing region list prices); Kimi K3 and K2.6; and Zhipu GLM-5.3 and GLM-5.3-Flash. US providers: OpenAI's GPT-6 family, Anthropic's Claude and Google's Gemini, from the same sources as the token counter. All prices are standard rates: no cache hits, no batch discounts, no promotions or free allowances. ByteDance's Doubao (Volcano Engine) pricing page only renders with JavaScript, so it could not be checked and is not included yet.

DeepSeek peak and off-peak

DeepSeek's off-peak price is half its peak price. Peak hours are 9:00–12:00 and 14:00–18:00 Beijing time, Monday to Friday, excluding Chinese public holidays; every other time, including weekends and public holidays, is off-peak. The tool uses the State Council holiday schedule from this site's China working days calculator, treating every official day off as off-peak, and treats Saturdays that are official make-up working days as off-peak, since they are not Monday to Friday. Running the same batch in the evening or at the weekend halves the DeepSeek bill.

Long-context pricing

Some models price by input length. When a GPT-6 family request has more than 272K input tokens, OpenAI bills the whole request at 2x for input and 1.5x for output; Gemini 3.1 Pro Preview also moves to a higher price above 200K. The table marks these cases. The Chinese models listed here have a single price at the rates shown.

Conversion and ranking

Each model's cost is first worked out in its own currency — input tokens × input price plus output tokens × output price, with prices per million tokens — then converted to your chosen currency and ranked. The default exchange rate is the central parity published by the China Foreign Exchange Trade System; change it to the rate you actually settle at.

Examples

Summarising a long document: 100,000 input tokens and 2,000 output tokens, compared in yuan at 6.7351 CNY per USD:

For the same job, the most expensive models in the table, GPT-6 Astra and Claude Fable 5.1 (about ¥7.41 each), cost 100 times as much as the cheapest, GPT-6 Luna. Light models suit bulk summarising and classification; save flagship models for hard reasoning and coding. Reasoning models bill their thinking as output, so real output is often longer than the visible answer — leave headroom.

Limits

• Only prices are compared. Capability, speed, reliability and rate limits differ a great deal; the cheapest is not necessarily the best fit.
• Context caching, batch discounts, promotions and free allowances are not included, and can cut real costs considerably.
• Prices change often; the table reflects the check date.
• Doubao and other models are not included yet.
• Currency conversion uses the rate you enter; each platform bills in its own way.

Sources and further reading

• DeepSeek: models and pricing (in Chinese)
• DeepSeek: token usage (in Chinese)
• Alibaba Cloud Model Studio: pricing (in Chinese)
• Kimi: model pricing (in Chinese)
• Zhipu AI: API pricing (in Chinese)
• OpenAI: API pricing
• Anthropic: Claude pricing
• Google: Gemini API pricing
• China Foreign Exchange Trade System: RMB central parity (in Chinese)