DeepSeek's API price adjustment officially took effect this week: cache-hit input rose from ¥0.025 to ¥0.3 per million tokens, a 1100% increase. We've noted that this Chinese LLM company, once dubbed the "price butcher," has finally put down the blade — and it likely won't be the last.

What this is

This hike only affects the API (Application Programming Interface — the channel letting external software directly invoke AI capabilities). Regular users chatting on web or App are completely unaffected. Flagship V4-Pro peak-hour (weekdays 9:00–12:00, 14:00–18:00) pricing:

  • Cache-hit input: 0.025→0.3 ¥ / million tokens, +1100%
  • Standard input: 3→9 ¥ / million tokens, +200%
  • Output: 6→27 ¥ / million tokens, +350%

Off-peak hours (evenings and weekends) are directly half-price, but even off-peak output costs ¥13.5 — still more than double compared to three months ago. Notably, just three months ago DeepSeek was loudly announcing a "permanent 75% price cut," and three months later it's reversed 11x.

Industry view

Put in the broader industry context, this move isn't surprising. Zhipu, Tencent Hunyuan, and Baidu Wenxin have all adjusted prices this year. Chinese LLMs have collectively shifted from "burning cash to grab market share" to "counting the bills to make ends meet." Compute costs are real: V4 is a trillion-parameter MoE (Mixture of Experts) model, and with Agents exploding in popularity, daily call volumes have skyrocketed. GPUs, electricity, ops — every line item is real money.

But our editorial team has also heard another voice: some developers argue the 11x hike is too steep, especially for cache-hit pricing — a design originally meant to encourage developers to reuse more — that overnight turned into a luxury. Small projects that built their cost models around the low prices now need to recalculate. Even more noteworthy: DeepSeek's playbook of using rock-bottom prices to pull competitors and users onto the platform, then locking them in with hikes, is itself more worth watching than the price numbers themselves.

Impact on regular people

  • For enterprise IT: Companies relying on DeepSeek API for document processing, intelligent customer service, and data cleaning need to re-budget. Off-peak scheduling and model tiering (using cheaper models for simple tasks) is now basic hygiene.
  • For individual professionals: Regular white-collar users on web or App are unaffected. But if your company embeds AI into business processes, this news hits the company ledger, not your workflow.
  • For consumer market: End-consumer (C-end) product prices won't change in the short term, but AI startups that can't absorb cost increases may shrink some free features or convert them to paid.