DeepSeek V4 Pro Raises Prices: How to Redo the Budget in One Evening

DeepSeek has released the V4 Pro 0813 model and, at the same time, raised API rates by roughly 1.5–2.5 times. Cache went up the most. During peak hours, a double multiplier applies on top of the rate.

What actually got more expensive — the request, or the habit of saving on it?

In quality, V4 Pro now sits level with or slightly below GPT-5.6 Luna, which is more widely available. The increase is uneven: base positions rose at the low end of the range, cache at the high end. The peak window doubles the bill on top of the new rate. The old math rested on cheap cache and direct availability from Russia — both supports have shifted.

Does the recalculation take an evening or a quarter?

A monthly inference budget is recalculated from actuals. Start with last month's usage export. Then calculate the share of tokens served from cache and the share of calls landing in the peak window. That shows how much of the increase can be removed by scheduling and how much by trimming the prompt. The winners are those whose context is assembled for the task rather than sent whole. To get a sense of the new order of magnitude, use the inference bill calculator.

The boundaries of the peak window and the exact per-position rates have not yet been named by the provider. Nor has the end date for old pricing for existing customers, payment terms, or access from Russia. Until those details arrive, budgets should be calculated at the top of the range.