DeepSeek warns of a 'significant' API price hike, no date yet
The cheapest frontier inference is about to get pricier: the second reversal in a month, and the China price war DeepSeek started is now unwinding.
Copy markdown
No number, no date — but 'substantial'
DeepSeek posted a notice saying its API prices will rise soon and the increase could be substantial, without giving new rates or an effective date. If you route production traffic through it for cheap inference, plan for higher bills and watch for the follow-up rate card.
What you pay now: $0.14 / $0.28 per million
V4-Flash currently runs $0.14 per million input tokens and $0.28 per million output — the floor that made DeepSeek the default cheap workhorse. The company hasn't said whether the hike hits V4-Flash, V4-Pro, cache-hit pricing, and reasoning models equally.
Second reversal in a month — surge pricing came first
In mid-July DeepSeek added peak/off-peak rates, charging more during busy hours; this warning is its second retreat from cheap-AI pricing in weeks. Together they read as a pivot from land-grab pricing to protecting margins.
The price war it started is unwinding
DeepSeek's permanent V4 discounts in May pushed ByteDance and Tencent to cut their own API prices; now the instigator is backing off. When the cheapest player blinks, the whole Chinese-model price floor tends to lift with it.
Still likely cheaper than US labs — for now
Even after the increase DeepSeek may stay cheaper than GPT-5.6 or Claude for many jobs, but the gap narrows. Worth benchmarking alternatives — Qwen3.8, self-hosted Kimi K3, GLM — and locking budgets before the new rates land.