deepseek-v4-flash pricing
Per million tokens, in US dollars. Synced from the public price feeds on 2026-08-07, and these are the same numbers KostLens bills from.
What it costs
| Provider | Input | Output | Cached input | Cache write | Cache write, 1h |
|---|---|---|---|---|---|
| deepseek | $0.140 | $0.280 | $0.003 | $0.000 | - |
| fireworks | $0.140 | $0.280 | $0.028 | - | - |
| tencent | $0.140 | $0.280 | $0.003 | $0.000 | - |
| azure | $0.190 | $0.510 | - | - | - |
| libertai | $0.250 | $1.75 | - | - | - |
A dash means the provider does not publish that rate, not that it is free.
The same model is not the same price
deepseek-v4-flash is sold by 5 providers here, and the dearest charges 79% more per input token than the cheapest: $0.250 on libertai against $0.140 on deepseek. That gap is real money on a workload of any size, and it is invisible in a provider console, which only ever shows you its own bill.
The rate most price tables leave out
Reading from the prompt cache is cheap, and everybody publishes that number. WRITING to it is a separate, disjoint set of tokens billed above the input rate, and almost nobody publishes it. On deepseek, deepseek-v4-flash charges $0.140 for fresh input and $0.000 to write into the cache.
It matters because a cost tool that folds cache writes into plain input understates the bill of every heavy cache user, and the people who reuse context are exactly the ones caching the most. KostLens counts them apart.
Know what this costs you, per customer
These are list prices. What they turn into on your bill depends on your token mix, and which of your customers is generating it. KostLens wraps your existing client in three lines, reads the usage the provider already returns, and gives you cost per customer and per feature. Your prompts and API keys never touch our servers.