DeepSeek
DeepSeek API
DeepSeek V4 — frontier reasoning at a fraction of Western frontier token prices.
2
моделей в линейке
DeepSeek V4 comes in two tiers, and the reason to look at them is price. Pro is a frontier reasoning model at an input price below most other vendors' cheap tiers, and Flash sits an order of magnitude below that again. Cache reads are the standout number: on Flash they are priced at roughly a three-hundredth of fresh input, the most aggressive caching discount in the catalog.
Коротко
- От
- 0.23 coins / 1 млн токенов
Модели этого семейства
| Модель | От | Режимы | Открыть |
|---|---|---|---|
| DeepSeek V4 FlashFast, very low cost DeepSeek tier for high-volume classification, extraction and chat. | 0.23 coins/ 1 млн токенов | — | ОткрытьДокументация |
| DeepSeek V4 ProDeepSeek's frontier reasoning model. Strong at maths, coding and long analytical chains at a fraction of Western frontier pricing. | 0.6 coins/ 1 млн токенов | — | ОткрытьДокументация |
Цены
| Модель | Вариант | Цена |
|---|---|---|
| DeepSeek V4 Flash | input | 0.23 coins/ 1 млн токенов |
| DeepSeek V4 Flash | output | 0.691 coins/ 1 млн токенов |
| DeepSeek V4 Flash | cacheRead | 0.0008 coins/ 1 млн токенов |
| DeepSeek V4 Pro | input | 0.6 coins/ 1 млн токенов |
| DeepSeek V4 Pro | output | 1.8 coins/ 1 млн токенов |
| DeepSeek V4 Pro | cacheRead | 0.02 coins/ 1 млн токенов |
Что выбрать
DeepSeek V4 Flash
High-volume classification, extraction and chat where per-token cost dominates the decision.
DeepSeek V4 Pro
Maths, coding and long analytical chains — frontier reasoning without frontier pricing.
Сильные стороны
- The cheapest frontier-class reasoning in the catalog on a per-token basis.
- Cache reads on Flash are priced at a rounding error, which makes long fixed prompts almost free to reuse.
- Output is only three times input on both tiers, a narrower ratio than the GPT or Claude lines.
- OpenAI-protocol compatible, so it drops into existing code.
Ограничения
- Only two tiers, with no mid-range option between them.
- Cache writes are not priced separately here, so caching economics differ from the GPT line.
- Text-only — no image input in this family.
Описание линейки от самого вендора: DeepSeek — API docs