DeepSeek V3.2 Pricing Update
A lot of older cost comparison content still describes DeepSeek as if V3 and R1 live in visibly separate price bands. The current official pricing page tells a cleaner story.
Current pricing snapshot
Using the official cache-miss input rate from DeepSeek's current docs:
- DeepSeek V3.2 / deepseek-chat: $0.28 input / $0.42 output per 1M tokens
- DeepSeek Reasoner / deepseek-reasoner: $0.28 input / $0.42 output per 1M tokens
The docs also publish lower pricing for cache-hit input tokens, which means the real optimization lever is not just model choice. It is prompt reuse and cache strategy.
Why this is important
This changes the buying conversation in two ways:
1. Label alone is less important
If the pricing baseline is shared, the decision between chat and reasoner modes becomes more about output quality, latency, and workflow fit.
2. Caching matters more
DeepSeek is one of the clearest examples of why API buyers should not stop at the headline token price. Once cache pricing becomes part of the official story, the actual savings opportunity shifts toward prompt engineering and repeated context reuse.
Practical takeaway
If you evaluate DeepSeek only by the older "cheap base model vs expensive reasoning model" framing, you may be using outdated mental models. The current cost conversation is more nuanced:
- Compare on quality and latency first.
- Compare on cache behavior second.
- Compare on base input/output price only after that.
Source
- DeepSeek API Pricing: https://api-docs.deepseek.com/quick_start/pricing