DeepSeek-V4-Pro vs DeepSeek-V4-Flash
Side by side on the numbers both labs publish: what they cost, how much they hold, when they shipped. No benchmarks, because we haven't run any.
From DeepSeek's and DeepSeek's own docs · updated 2026-09-05 10:32 UTC
- DeepSeek-V4-Pro charges 3x what DeepSeek-V4-Flash does for the words it writes back.
- At steady daily use, DeepSeek-V4-Flash works out around $0.77 a month and DeepSeek-V4-Pro around $2.32.
- DeepSeek-V4-Pro is the newer of the two, out on 2026-08-13 against 2026-07-31.
| DeepSeek-V4-Pro | DeepSeek-V4-Flash | |
|---|---|---|
| Lab | DeepSeek | DeepSeek |
| Released | 2026-08-13 | 2026-07-31 |
| Status | released | released |
| Built in | China | China |
| Context window | 1M | 1M |
| Longest reply | 384K | 384K |
| Input, per million | $1.32 | $0.44 |
| Output, per million | $3.96 | $1.32 |
| API name | deepseek-v4-pro | deepseek-v4-flash |
| A few questions a day, a month | $0.42 | $0.14 |
| Steady daily use, a month | $2.32 | $0.77 |
| Something running all day, a month | $54 | $18 |
The dot marks the cheaper price, the bigger window or the longer reply. Monthly costs assume the workload in the row, thirty days, list prices, no caching or batch discounts. Price every model at once.
What this comparison can't tell you
Which one writes better. Which one is easier to steer. Which one your own prompts already work on. Those depend on your work, and a spec table has never answered them.
What it does settle is the money and the ceiling. If two models both clear your bar, the table tells you which one is cheaper to live with, and by how much.
Where the numbers came from
DeepSeek-V4-Pro: DeepSeek documentation. DeepSeek-V4-Flash: DeepSeek documentation. We refetch both with the rest of the pipeline, so the stamp at the top says when we last looked.
Compare two others · everything on DeepSeek-V4-Pro · everything on DeepSeek-V4-Flash