Support retrieval
High cache reuse and short answers. Track cache-hit input separately; an output cap often matters more than request count.
Predictable DeepSeek V4 Budgeting — No Peak-Hour Surprises
Compare DeepSeek's official time-based list prices with AIWave's published all-day rates, then enter your own token volumes. The calculator makes the trade-off visible: official list prices are lower, while a unified rate removes the peak-window variable and keeps multiple Chinese model families behind one API.
Model a monthly billTwo different pricing structures
Important: official DeepSeek prices below use peak and off-peak windows. AIWave prices are separate all-day rates. AIWave is above the official peak list price; choose it for budget structure and gateway workflow only when those benefits fit your workload.
| Route and window | Model | Input / 1M | Output / 1M | Cache hit / 1M |
|---|---|---|---|---|
| DeepSeek official · peak | V4 Flash | $0.440 | $1.320 | $0.0140 |
| DeepSeek official · off-peak | V4 Flash | $0.220 | $0.660 | $0.0070 |
| AIWave · all day | V4 Flash | $0.638 | $1.914 | $0.0203 |
| DeepSeek official · peak | V4 Pro | $1.320 | $3.960 | $0.0440 |
| DeepSeek official · off-peak | V4 Pro | $0.660 | $1.980 | $0.0220 |
| AIWave · all day | V4 Pro | $1.914 | $5.742 | $0.0638 |
How the official peak windows work
DeepSeek's official schedule uses Beijing time. The peak windows are 09:00–12:00 and 14:00–18:00; listed peak rates are twice the off-peak rates. A batch that moves across a window boundary therefore needs a time-share assumption. The AIWave rate card on this page uses one rate throughout the day, so its calculator does not need a clock-based multiplier.
Monthly token budget calculator
Official schedule estimate$0.00Peak/off-peak weighted
AIWave all-day estimate$0.00No time-share input
Difference$0.00AIWave minus official estimate
Three workload scenarios
Scheduled batch extraction
A controllable schedule can exploit official off-peak pricing. An all-day rate is easier to forecast, but that predictability has a higher listed token cost.
Interactive reasoning
Traffic arrives when users are active and output varies. Use Pro only where an acceptance test shows a quality benefit, then set retry and output budgets.
Use the comparison responsibly
Rate cards change. Preserve the rate date, model slug, cache definition, token counts, currency, retries, and provider route with every forecast. Do not infer latency or quality from price. Test representative prompts and compare completed-task cost before choosing a route.
For the current AIWave catalog, see pricing and the model directory. For implementation details, use the chat completions guide.