DeepSeek V4 Pro
The larger DeepSeek still on the official API. Index 36 at max, behind its own flash model.
Open weightsPeak rate shown
36Index · Max
$0.67Per measured task
53.7 pts/$Points / $
81Tokens / sec
No High score is published. Showing Max.
Every published effort
| Setting | Index | Per task | Tokens / sec |
|---|---|---|---|
| Max | 36 | $0.67 | 81 |
| Published | 20 | $0.48 | 180 |
What this workload costs
List price for one call. Reasoning tokens count as output. A cache hit uses the lab’s cache rate when one is published.
DeepSeek V4 Pro
Cache appliedSpecs
- API id
- deepseek-v4-pro
- Released
- 13 Aug 2026
- Context
- 1M
- Max output
- 384k
- Input / 1M
- $1.32
- Output / 1M
- $3.96
- Cache read / 1M
- $0.04
- Knowledge
- —
- Inputs
- Text
- Weights
- Open weights. Confirm the license on the 0813 card.
Peak cache-miss rate for the 0813 checkpoint. Off-peak is half. The date is the 0813 checkpoint name.
The pricing write-up lists this checkpoint as text, without vision.
Use it when
You specifically need the Pro checkpoint. On this index, V4.1 Flash scores higher and costs less per measured task.
Leave it when
Treating Pro as the automatic upgrade over Flash. In this snapshot it is not.