DeepSeek's Vague 'Significant' Price Hike Leaves Developers Guessing

I just received an email push notification with DeepSeek's price adjustment notice. The wording was very polite: "We plan to raise API service pricing overall in the near future, and the increase is expected to be significant. Please arrange your usage reasonably. If it's not suitable, you can still get a refund."

The exact amount of the increase wasn't stated, but it emphasized that the "increase is significant".

Honestly, I've been using DeepSeek-V4-Pro and Flash a lot recently, mainly connected to Trae and Codex as my primary models.

A few days ago, after the official version of the DeepSeek-V4-Flash API went into public beta with significantly enhanced Agent capabilities, it became the model I use most for daily coding and running tasks. So this notification is not a small matter for me. With a "significant" price increase, I really have to think it over carefully.

The price increase itself doesn't surprise me

For any model service, you eventually have to pay back what you've freeloaded. DeepSeek entered the market with extremely low prices, scaling up volume first before talking about profitability. This is a very normal path for a commercial company.

That said, if you truly believe it "should always be cheap," that's naive. Compute power, bandwidth, and inference costs are all real. Long-term loss-making is unsustainable.

What really makes me sigh is the "vagueness" of this notice. We don't know how much it will increase or when. It only leaves a line: "The increase is expected to be significant, subject to the official notice."

For individual developers, this uncertainty is the most painful part—you rely on it to run automated tasks and tests, and suddenly one day your bill doubles, forcing you to recalculate project costs.

What I care more about is "Is it worth it?"

Whether a price increase is acceptable depends on whether the value has kept pace. The Agent capabilities of this generation's V4-Flash are a real improvement. After connecting it to Codex, the experience of "just give it a requirement and it runs on its own" feels much smoother than before (though still not comparable to GPT, but it's not bad). If the price increase matches the actual capability improvement, I believe many people will be willing to continue using it.

What's worrying is a product where the "price rises faster than capability." Tools can raise prices, but users need to feel the money is well spent, not that they're being held hostage.

Trae also recently adjusted its strategy, automatically converting my previous speed-pass credits into points! A forced switch.

My 70 speed-pass uses were converted to roughly 2000 points, plus a veteran user benefit!

The only downside is that speed-passes could be enabled on demand, whereas points are different—conversations consume them continuously, and once you run out of points, you can't use the service anymore. Trae can no longer be used for free going forward.

However, Trae Work also supports DeepSeek's related models, and compared to other models, it is indeed the cheapest.

Various model vendors and tools are all tightening up. What does this signal?

My coping strategy

1. Diversify dependencies. DeepSeek, other domestic alternatives, and running open-source models locally. Many people treat a single model as their main workhorse, but don't stake your lifeline on just one.
2. Pay attention to open-source weights. As model capabilities improve, local deployment for private solutions is a long-term fallback for many teams.
3. Re-evaluate usage. Cut down on high-frequency, low-value calls. If something can be handled with caching and rule-based processing, don't spend money calling the model every time.

A final couple of words

The price increase notice itself is nothing special—just a business move. Perhaps it will become harder and harder for us to freeload, and we'll truly be paying to work.

The model is good, I acknowledge that; a reasonable price increase, I can also accept. I just hope that when the official plan comes out, they can give a decent reason, and not leave everyone paying in confusion.

After all, tools are meant to be used, not worried about. I hope the increased fees will be invested in computing power and R&D, so they can truly go toe-to-toe with foreign models. That would be perfect.