DeepSeek Flash vs V4 Pro

Same 1M context. The differences are price, vision and reasoning depth.

	Flash	V4 Pro
Input, cache miss (off-peak)	$0.15	$0.66
Output (off-peak)	$0.60	$1.98
Price multiple	1x	3.3–4.4x
Context	1M	1M
Max output	384K	384K
Vision input	Yes	No
Max concurrency	2500	500

Where Flash wins outright

Vision tasks. V4 Pro does not accept images. If your input is a screenshot or a scan, Flash is the only option.
High concurrency. 2500 versus 500 concurrent requests — five times the throughput ceiling.
Volume classification, extraction, translation. Pattern work where the gap in quality is small and the gap in price is 3–4x.

Where V4 Pro is worth the multiple

Code review and debugging. A missed bug costs far more than the tokens.
Multi-step reasoning. Analysis, trade-off decisions, anything where Flash tends to stop at the first plausible answer.
Long-form professional writing. Structure and consistency degrade noticeably when you cut model capability.

The practical rule

Ask one question: what does a wrong answer cost?
Under a cent — use Flash, and spend the savings on volume.
More than the token cost — use V4 Pro, and stop optimising pennies.

You can mix them

Most production systems use both: Flash for the high-volume path (classification, routing, extraction), V4 Pro for the low-volume path where quality matters (final generation, review, judgement). Route by task, not by user.

Calculate it

Cost calculator — enter your token volume and compare both models directly.