【緊急解説】DeepSeek衝撃の値上げ発表！最新AIモデル価格徹底比較DeepSeek has announced a significant price increase for its DeepSeek-V4 API…

DeepSeek, the Chinese AI startup that built its global reputation on aggressively low pricing, has announced a significant price increase for its flagship DeepSeek-V4 API, along with the introduction of dynamic peak-time pricing. The shift matters because DeepSeek had positioned itself as the industry's cost leader, and any change to its rates ripples through the calculations of developers, startups, and enterprises that adopted its models specifically to keep inference costs down.

For much of the past two years, DeepSeek was known less for headline benchmark scores than for undercutting nearly every competitor on price. Its earlier releases were widely described as a "price destroyer," offering per-token rates that were a fraction of what leading Western labs charged for comparable capability. That combination of open weights and rock-bottom API pricing was central to the original "DeepSeek Shock," when the company's models forced the market to reassess how cheaply frontier-adjacent performance could be delivered. The latest announcement, which the source frames as a potential second chapter of that shock, appears to reverse that dynamic by moving toward what the company presents as price normalization rather than continued price cutting.

The two headline changes are a substantial increase to the standard DeepSeek-V4 API rate and a new dynamic pricing model tied to peak usage periods. Peak-time or time-of-day pricing is a familiar concept in cloud computing and utilities, but it is still relatively uncommon in the large language model API market, where flat per-token rates have been the norm. Under such a scheme, requests sent during high-demand windows would likely cost more than those sent during off-peak hours, giving customers a financial incentive to schedule batch workloads, background jobs, or non-urgent processing outside busy periods. For interactive, latency-sensitive applications that cannot easily shift their traffic, the effective cost increase could be more pronounced.

Several factors could plausibly explain the move, though DeepSeek's precise reasoning is not fully detailed in the announcement. Serving a frontier-scale model at very low prices is expensive, and sustained demand can strain GPU capacity, particularly given the constraints Chinese firms face around access to the most advanced accelerators. Introducing peak-time pricing is a common way to smooth demand and manage infrastructure load, while a base price increase may reflect a push toward profitability or sustainability after a prolonged period of loss-leading rates. It is reasonable to read the change as a maturing business trying to balance growth against the real costs of operating at scale, rather than as a retreat from the market.

The broader context is a market where API pricing has been trending downward for most providers. OpenAI, Google, and Anthropic have repeatedly cut prices or released smaller, cheaper model tiers such as mini and flash variants, while open-weight alternatives from Meta's Llama family, Alibaba's Qwen, Mistral, and others give developers the option to self-host and avoid per-token API fees entirely. Against that backdrop, a price increase from the provider best known for being the cheapest is notable, and it may narrow the gap that made DeepSeek so attractive relative to established competitors. Whether rivals interpret this as room to hold their own prices, or whether the overall downward trend continues, is likely to become clearer as customers respond.

For teams evaluating options, the practical takeaway is that comparing API prices now requires looking beyond a single advertised per-token figure. Dynamic pricing means total cost depends on when and how workloads run, so straightforward comparisons across providers become more complex. Because DeepSeek publishes open weights for many of its models, organizations with sufficient hardware and engineering capacity retain the alternative of running the models locally or through third-party hosts, which is one reason this development sits within the local and open-model conversation rather than being purely a cloud pricing story.

The article this summary is drawn from compiles current API rates across major AI models to place DeepSeek's revised pricing in perspective and to assess the potential market impact. As with any single announcement, the longer-term effect will depend on how DeepSeek's users, competitors, and the wider ecosystem adjust, and readers should treat early reactions as provisional until adoption patterns and competitor responses settle.