DeepSeek API Prices Jump 11× Overnight, Ending the Subsidized Era

I opened the DeepSeek backend and saw that the new pricing has officially taken effect.

The most brutal change is the cache-hit input price, which jumped from 0.025 yuan to 0.3 yuan per million tokens — an increase of 1100%.

That's right, 11 times.

The "price butcher" that once drove large-model pricing into the ground has finally sheathed its knife.

Let's do the math: how much did it actually go up?

This price adjustment only affects the API interface; regular users chatting via the web or app are completely unaffected.

But for developers and small teams who rely on the API, the impact is significant.

Flagship model V4-Pro during peak hours:

• Cache-hit input: 0.025 → 0.3 yuan, up 1100%
• Standard input: 3 → 9 yuan, up 200%
• Output: 6 → 27 yuan, up 350%

Peak hours are weekdays 9:00-12:00 and 14:00-18:00. The rest of the time is considered off-peak, at half price.

In other words, calling the API in the evening or on weekends saves half the cost.

But even at the off-peak half-price rate, it's still much more expensive than before the adjustment.

Previously, V4-Pro output cost only 6 yuan per million tokens; now it's 13.5 off-peak and 27 during peak hours.

Why such a sudden, steep increase?

To put it bluntly: the cheap days were always subsidized.

Three months ago, DeepSeek officially announced a permanent 75% price cut, driving the cache-hit price down to a rock-bottom 0.025 yuan.

At that time, everyone was shouting that domestic large models were in a crazy price war.

But computing power costs are real.

V4 is a trillion-parameter MoE architecture. With the recent explosion of AI Agents, daily call volumes have surged. GPUs, electricity, and maintenance all cost money.

Previously, they burned cash to grab market share. Now that they have it, they need to return to commercial fundamentals.

Plus, the whole industry has been raising prices this year; Zhipu, Hunyuan, and Wenxin all adjusted their prices earlier.

DeepSeek's move is really just catching up with the pack.

Some personal thoughts

I have a few small tools of my own that have always used DeepSeek's API.

Honestly, I chose it initially because it was cheap.

With the same performance, it was more than half the price of competitors. I could use it freely without worrying about the cost.

Batch processing documents or running Agent tasks, I could go wild, and it only cost a few dozen yuan a month.

Last night, I deliberately rushed to finish a batch of tasks before the price change took effect.

Calculating the cost today, the same workload would cost two to three times more during peak hours.

It would be a lie to say it doesn't hurt.

But thinking about it calmly, I can understand.

There's no such thing as a permanent free lunch.

A tech company needs to sustain R&D and can't keep selling at a loss forever.

It's just that the magnitude of this increase was sudden, and the cost models for many small projects have been thrown into disarray.

A few suggestions for regular developers

First, shift workloads off-peak whenever possible.

Move non-urgent batch tasks to evenings or weekends to save half the cost immediately.

Keep only real-time interactive interfaces during peak hours to significantly reduce costs.

Second, choose models based on the scenario.

For simple tasks, V4-Flash is sufficient. Its peak output cost is 9 yuan per million tokens, which is two-thirds cheaper than Pro.

Not every scenario requires the top-tier model.

Third, optimize call logic.

Cache everything that can be cached to reduce redundant inputs.

Break down long-text tasks; there's no need to feed the entire document every time.

Fourth, keep multiple options ready.

The price gap between different providers is narrowing now. Test several to find the best fit for your specific scenario.

Don't put all your eggs in one basket.

A final honest word

DeepSeek's price increase marks the official end of the dirt-cheap era for domestic large models.

For users, this is certainly not good news.

But for the industry as a whole, it might not be a bad thing.

If the price war continued, no one would have money for R&D, and ultimately, no one could produce a good product.

Reasonable profits are necessary to support sustained technological progress.

It's just tough for us developers who got used to the low prices; we have to recalculate our budgets again.

Life goes on, and we'll still use what we need to use. We'll just have to be a bit more frugal with our API calls from now on.