DeepSeek V4 Pro
The reasoning model. Three to four times Flash's price; worth it only for the hard cases.
Input (off-peak)
$0.66
Output (off-peak)
$1.98
Context
1M
Concurrency
500
Specifications
| Property | Value |
|---|---|
| Model id | deepseek-v4-pro |
| Context window | 1,000,000 tokens |
| Max output | 384,000 tokens |
| Vision input | No |
| Max concurrency | 500 |
| Input, cache hit (off-peak) | $0.022 / 1M |
| Input, cache miss (off-peak) | $0.66 / 1M |
| Output (off-peak) | $1.98 / 1M |
| Peak multiplier | 2.0x |
What it is good at
- Code review, debugging, refactoring
- Multi-step analysis and trade-off decisions
- Contract and document interpretation
- Long-form professional writing
- Anything where the answer must survive scrutiny
No vision
V4 Pro does not accept image input. If your pipeline includes screenshots,
scanned documents or diagrams, you need Flash for that step —
either instead of, or before, V4 Pro.
Provider spread is smaller
Only 16 providers resell V4 Pro, with a spread of about 2.2x — far narrower than Flash's 18x. That is because serving a large reasoning model profitably requires serious hardware. See all V4 Pro quotes.
When it is not worth it
If your task is classification, extraction, translation or formatting, V4 Pro costs three to four times more to do the same job. Use the model picker.