An email that made developers stop and read twice
The notice arrived in inboxes this week, addressed to registered API users: "We plan to raise the overall pricing for DeepSeek API services in the near future, with a significant increase expected. Please plan your usage accordingly." The company adds that continued use after the adjustment will be treated as acceptance of the new terms — and that those who disagree can cancel and request a refund.
The same warning now sits directly on the official pricing page as a footnote under the price table, which means this is a formal announcement, not a rumour. The story was quickly picked up by TechNode and Dataconomy, citing a Bloomberg report. What nobody knows yet: the size of the increase, the effective date, and whether it applies equally to every model.
Why this hits so hard: DeepSeek has no real rival on price
To understand the stakes, look at what DeepSeek charges today. DeepSeek V4 Flash costs 0.14 USD per million input tokens and 0.28 USD per million output tokens. The more powerful V4 Pro runs at 0.435 USD input and 0.87 USD output. Both models offer a 1-million-token context window.
Compare that with the flagship competition: Claude Opus 4.8 costs 5 USD input and 25 USD output per million tokens; OpenAI's GPT-5.6 Sol runs at 5 USD and 30 USD. V4 Flash is roughly 35 times cheaper on input and 90 times cheaper on output than the top Western models — and it is not a toy. With 284 billion parameters, an MIT license and agentic coding scores that come within a few points of Claude Opus 4.8, it became the model European developers could actually afford to build with. Last week it topped OpenRouter's weekly ranking with 7.22 trillion tokens processed.
If you are looking for an alternative at that price-performance point, there effectively is none. That is precisely why the announcement matters.
The twist: OpenAI just made its own models cheaper
Here is where the story gets interesting. On 30 July 2026, OpenAI reduced the price of GPT-5.6 Luna by 80 percent and GPT-5.6 Terra by 20 percent, according to the official GPT-5.6 announcement. Luna, OpenAI's most cost-efficient model, now costs 0.20 USD input and 1.20 USD output per million tokens — down from 1.00 and 6.00 USD at launch. Terra dropped from 2.50/15.00 USD to 2.00/12.00 USD.
Now compare Luna with the original DeepSeek price list. When DeepSeek V4 launched in April 2026, V4 Pro was officially priced at 1.74 USD input and 3.48 USD output per million tokens — roughly eight times the price it charges today. The current ultra-low rates came later, in May, when DeepSeek announced a permanent discount and cut cache-hit prices to a tenth of the launch level.
Put those numbers side by side:
| Model | Input (USD/1M) | Output (USD/1M) |
|---|---|---|
| DeepSeek V4 Flash (today) | $0.14 | $0.28 |
| DeepSeek V4 Pro (today, after May discount) | $0.435 | $0.87 |
| DeepSeek V4 Pro (original launch price) | $1.74 | $3.48 |
| GPT-5.6 Luna (after 80 % cut) | $0.20 | $1.20 |
| GPT-5.6 Luna (original) | $1.00 | $6.00 |
The implication is clear. DeepSeek first published a price list, then slashed it, and now announces a significant increase. The most plausible reading is a return toward the originally published level. And if that happens, GPT-5.6 Luna at 0.20/1.20 USD suddenly becomes a genuine competitor — something that was unthinkable when DeepSeek's promotional pricing ruled the market.
What this means for European developers and startups
For European startups and freelancers, DeepSeek's pricing has been a quiet gift: a frontier-level model you could run for pocket change, no credit card gymnastics, no per-seat subscriptions. A million input tokens cost about 0.12 EUR at today's rates — less than a cup of coffee. That is the kind of arithmetic that let small teams ship AI features they could never have justified with US pricing.
The practical advice is simple. Watch the Open Platform announcements for the new price list. Recalculate your token budgets and reserves before the change lands. And remember the European option that US developers often overlook: because V4 Flash is open-weight under an MIT license, you can run it on your own hardware — data stays in the EU, no per-token bill, no dependency on a Chinese API. We looked at the cost of local deployment in our earlier comparison of budget models, where cheap gets expensive.
One thing is certain: the era of the 0.14-dollar frontier model may be ending. The smart European teams are already planning for both scenarios — a mild increase they can absorb, and a return to original pricing that would finally hand OpenAI's Luna a real opening.
When will DeepSeek publish the new prices?
DeepSeek has not given a date or figures — the company only says the plan will follow via official notice on the Open Platform and email. Neither TechNode nor Bloomberg could confirm a new price list or an effective date.
Will the increase apply to every DeepSeek model equally?
Not stated. DeepSeek has not said whether V4 Flash, V4 Pro, cache-hit pricing and reasoning-focused models will all rise by the same amount — that detail will only come with the official notice.
Is GPT-5.6 Luna a real alternative for European developers?
After the 80 percent price cut, Luna at 0.20/1.20 USD is the closest thing to a DeepSeek-style bargain from a US provider — especially if DeepSeek returns to its original 1.74/3.48 USD level for V4 Pro. It is available in the EU through the standard OpenAI API.
