DeepSeek says it plans to implement substantial price increases across its AI services; V4 Flash currently costs $0.14/1M input and $0.28/1M output tokens
DeepSeek plans to implement a significant price increase across its AI services, an unusual shift from the disruptive Chinese player that has put pressure on US and domestic rivals.
Bloomberg
Context & Ripple Effects
DeepSeek had made aggressive API pricing a central part of its positioning: in May it made a 75% V4 Pro price cut permanent, while V4 Flash was already listed at $0.14 per million input tokens and $0.28 per million output tokens. The planned increase reverses that recent direction.
The move follows an estimate that V4-Flash averaged $0.03 per test, far below several named rivals, underscoring why a change in DeepSeek's pricing matters beyond its own customer base.
First-order effects
DeepSeek API customers will face higher service costs when the planned increases take effect, including users of V4 Flash at its currently published token rates.
DeepSeek shifts from extending its earlier permanent Pro discount to raising prices across its AI services, changing the commercial terms confronting existing customers.
Second-order effects
US and Chinese rivals that have been pressured by DeepSeek's low prices gain more room to defend their own API pricing rather than matching its prior levels.
Customers comparing models on cost per task will need to recalculate DeepSeek's advantage once the company publishes new rates, potentially changing workload-routing decisions.
Third-order effects
If major low-cost model providers increasingly revise prices upward after using discounts to win adoption, AI API competition will be shaped less by headline token prices and more by sustainable inference economics and effective task cost.
The reversal points toward more capacity-aware pricing in AI services, where providers balance low entry prices against the cost of serving growing usage.
The trend: AI model providers are moving from price-led customer acquisition toward pricing that more directly reflects the economics of operating inference services.
on the upcoming deepseek price increase we've been able to reproduce their current prices even on rented GPUs so this likely isn't because they're “losing money” it's traffic shaping because they are overloaded
DeepSeek is now sending out mails and offering refunds in regards to this “price hike” they announced with a banner on their API Console. (they did this with the peak hours pricing announcement too) There's no way they go upto Kimi K3 prices... right? [image]
They don't have enough inference capacity to fulfil demand, so they are trying to raise prices and destroy demand One of the reasons why I'm not worried about closed-source lab revenue is that they or their investors (hyperscalers like Google, Amazon, Microsoft) control the [imag…
DeepSeek: «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» Oof. They originally promised price *cuts* in H2. given the market situation... they are heavily overloaded. [im…
Chinese LLMs may not end up being so much cheaper than rivals. The worry: China will keep flooding the market with cheap LLMs to gain market share. The reality: investors in Chinese companies are demanding greater profitability before giving more money. https://www.bloomberg.com/…
Oh no can we still call them cost curve rebels?? I'm going to venture and bet that this price increase has a very good reason behind it and is not about just gouging profit though
DeepSeek's parent company High Flyer quant fund saw all nine funds lose more than 20% in July erasing YTD gains. I hope Liang wasn't on the same levered AI trades as Leopold. 😬 Regardless, I don't think that's why DS is raising prices on their models though. 🤣
deepseek's liang wenfeng on raising prices, from the leaked investor call “if we wanted to maximise profit, we would charge more. at this price point, demand is inelastic: if i raised the price by another 50 per cent, or even doubled it, token consumption would not change very [i…