DeepSeek Slashes API Prices Permanently to One-Quarter of Original

I was a bit busy over the weekend and only saw DeepSeek's permanent price cut while scrolling through my phone last night. At first glance, I missed the word "permanent."

Later, I saw it again and realized that the DeepSeek Pro model API price, after the 75% discount promotion ends on May 31, 2026, will officially be adjusted to one-quarter of the original price.

DeepSeek's copywriting is quite humorous—a 75% discount becomes one-quarter.

As one netizen put it, "The limited-time offer ended, and I thought what ended was the discount, but it turned out what ended was the 'limited-time' part."

To those who were hammering V4 on the first day, you are so lucky—you got to experience V4 at the limited-time original price~

But then again, I think I also tested it for a little over 1 yuan at the time.

Even though I lost a few cents more, I'm still happy. After all, the price butcher of large models has awakened again, and there's no need to keep rationing tokens from now on.

Besides the extremely low unit price, DeepSeek has another stunning feature—KV caching.

I've been using the DeepSeek API quite a bit recently, and after several tests, the cache hit rate actually reached 98%. I haven't even used the various tools people talk about online, and it's said the rate can go even higher.

Maybe a dry percentage isn't very intuitive, so let's calculate a price comparison with an example.

My recent daily token consumption is around 5 million. Let's assume 70% input and 30% output.

Before the price cut (no cache)

Daily cost = 5M × 70% × 12 yuan/million tokens + 5M × 30% × 24 yuan/million tokens = 42 + 36 = 78 yuan/day

After the price cut (no cache)

Daily cost = 5M × 70% × 3 yuan/million tokens + 5M × 30% × 6 yuan/million tokens = 10.5 + 9 = 19.5 yuan/day

After the price cut + 98% cache hit rate

Daily cost = 5M × 70% × 98% × 0.025 yuan/million tokens + 5M × 70% × 2% × 3 yuan/million tokens + 5M × 30% × 6 yuan/million tokens ≈ 0.09 + 0.21 + 9 = 9.3 yuan/day

Not to mention that other models' prices are even higher than DeepSeek's original price, and they also impose rate limits at the drop of a hat.

No wonder everyone calls him Saint Liang~

If you want to know what the long-term usage is like, wait until I've used it for a while and I'll report back to you all.