DeepSeek API Pricing & Model Comparison
We track 3 DeepSeek models, with input pricing from $0.14 to $0.43 per million tokens. The line-up's blended rate runs 86% below the catalog average. Below: price, context, speed and value compared across the range.
For cost, DeepSeek's cheapest option is deepseek-v4-flash ($0.14 input / $0.28 output); the quality ceiling is deepseek-reasoner-v4 at a quality score of 92. On a 50,000-turn-per-month support workload that is $21 versus $65 — the spread inside a single provider is often wider than the spread between providers.
On value (quality score ÷ output price) the pick of the DeepSeek range is deepseek-v4-flash. The largest context window is 1000K, offered only by deepseek-v4-flash, and the fastest output is deepseek-v4-flash at 180 tok/s. Modalities across the line-up: Text.
Models and pricing
Sorted by blended rate, cheapest first (3:1 input-to-output mix). Follow a model name for full specs and monthly cost estimates.
| Model | Input /1M | Output /1M | Blended /1M | Quality | Context | Speed | Value | Use cases |
|---|---|---|---|---|---|---|---|---|
| deepseek-v4-flash | $0.14 | $0.28 | $0.18 | 88 | 1000K | 180 tok/s | 314.3 | ChatbotAffordableFastLong Context |
| deepseek-chat-v4 | $0.26 | $1.03 | $0.45 | 88 | 131K | 140 tok/s | 85.5 | ChatbotAffordableFast |
| deepseek-reasoner-v4 | $0.43 | $0.87 | $0.54 | 92 | 131K | 30 tok/s | 105.7 | ReasoningCoding |
Frequently asked questions
Which DeepSeek model is cheapest?
deepseek-v4-flash, at $0.14 input and $0.28 output per million tokens, with a quality score of 88.
What is DeepSeek's flagship model?
By quality score it is deepseek-reasoner-v4 (92), priced at $0.43 input / $0.87 output with a 131K context window.
Does DeepSeek pricing change?
Yes. llmprice.app runs a daily collection job and refreshes these figures, but there can be a lag after a provider re-prices — confirm against DeepSeek's official pricing page before committing.
Prices and specs are collected daily by llmprice.app. Confirm against the official pricing page before production use.