DeepSeek slashes fees for new AI model in Chinese price war

```json { "title": "DeepSeek Slashes V4-Pro Fees 75% in AI Price War", "metaDescription": "DeepSeek cuts V4-Pro API fees by 75% until May 5, 2026, reigniting China's AI price war. What it means for developers and the global AI industry.", "content": "<h2>DeepSeek Cuts V4-Pro API Prices by 75%, Reigniting China's AI Price War</h2>\n\n<p>Hangzhou-based AI lab DeepSeek has slashed the cost of accessing its newly released DeepSeek-V4-Pro model by 75%, firing the latest shot in an escalating price war among Chinese artificial intelligence companies. The limited-time promotional discount, valid until May 5, 2026 at 15:59 UTC, applies to the V4-Pro's API pricing and comes alongside a permanent 90% reduction in input cache hit fees across all of DeepSeek's models. The moves have already rattled rival Chinese AI firms and drawn renewed global attention to the competitive dynamics reshaping the AI industry.</p>\n\n<h2>What's in the V4-Pro: Scale, Efficiency, and Rock-Bottom Pricing</h2>\n\n<p>DeepSeek released V4-Pro and its smaller sibling, V4-Flash, in preview on April 24, 2026—the company's first significant model launch since December 2025. V4-Pro is a substantial technical achievement on paper: it features 1.6 trillion total parameters with 49 billion active parameters per inference pass, making it the largest open-weight model currently available, according to Dataconomy. V4-Flash, by contrast, features a more compact 284 billion parameters.</p>\n\n<p>One of the model's most notable efficiency claims is its compute footprint. According to Dataconomy, V4-Pro requires only 27% of the computing power needed by its predecessor, V3.2, to handle a one-million-token context window. That context window—capable of processing the equivalent of several full-length novels in a single prompt—is paired with what DeepSeek describes as best-in-class performance among open-source models on agentic coding and reasoning tasks, according to Fortune.</p>\n\n<p>The pricing structure during the promotional period is striking by any measure. V4-Pro's standard rate is set at 3 yuan for input and 6 yuan for output per million tokens. The input cache hit price, however, has been slashed to just 0.025 yuan—approximately $0.0036—per million tokens. For context, OpenRouter data cited by Dataconomy puts Western competitors' output prices in the range of $12 to $25 per million tokens. Even outside of the promotional window, the permanent 10% pricing on cache hits represents a significant structural reduction for developers building on DeepSeek's infrastructure.</p>\n\n<p>The pricing changes were confirmed publicly by DeepSeek senior researcher Victor Chen on X, according to ZeroHedge. Chen clarified that while the 75% base discount runs only through May 5, the input cache hit reduction is permanent across all models.</p>\n\n<p>Huawei also confirmed that its latest AI computing cluster, powered by its Ascend AI processors, can support DeepSeek's V4 model, according to CNBC. That detail carries strategic weight: it signals that DeepSeek's most capable models can run on domestic Chinese hardware, reducing reliance on Nvidia chips that have been subject to U.S. export controls targeting China.</p>\n\n<h2>Rivals React, Markets Move, and the Price War Intensifies</h2>\n\n<p>The market response to DeepSeek's pricing announcement was swift and punishing for its Chinese competitors. According to Invezz, shares of MiniMax Group (HK:0100) dropped 10% in Hong Kong trading following the announcement, extending losses from the previous session, while Zhipu AI (Knowledge Atlas Tech, HK:2513) fell 3.4%. The selloff reflects a familiar pattern: when DeepSeek makes an aggressive move on cost, the competitive pressure on peers that cannot match those economics becomes immediately visible in their valuations.</p>\n\n<p>Bloomberg and Business Standard have both noted that DeepSeek's V4 pricing strategy threatens to re-ignite the broad AI price war that erupted in China following the release of the company's R1 model in January 2025. That earlier episode had consequences far beyond Chinese markets: according to Britannica, R1's release caused Nvidia's share price to fall 17%, wiping out nearly $600 billion in market capitalization in a single day—the largest one-day loss for a single company in U.S. stock market history.</p>\n\n<p>Whether V4-Pro produces a similar shock to global markets remains to be seen. DeepSeek has itself acknowledged a meaningful performance gap: according to Dataconomy, the company has stated that V4 models currently lag behind frontier models such as GPT-5.4 and Gemini 3.1 Pro by approximately three to six months in overall capability. That candid admission distinguishes V4-Pro's positioning from the R1 moment, when the model's performance relative to its reported development cost triggered a rapid reassessment of assumptions about AI's cost curve.</p>\n\n<h2>Context: Why DeepSeek's Pricing Strategy Matters Beyond China</h2>\n\n<p>DeepSeek is owned by Chinese hedge fund Zhejiang High-Flyer Asset Management, whose co-founder Liang Wenfeng formed the startup in 2023, according to Bloomberg. Unlike most of its peers, DeepSeek has operated without external venture funding—until now. According to PYMNTS, citing The Information, DeepSeek is in talks with Tencent and Alibaba about participating in the company's inaugural funding round, with DeepSeek seeking to raise more than $300 million at a valuation of at least $20 billion. Bloomberg reported separately that Tencent proposed acquiring up to a 20% stake in DeepSeek as part of the financing round, though the startup was reportedly reluctant to cede that level of control.</p>\n\n<p>The funding conversations are unfolding against a backdrop of escalating investment across the Chinese AI sector more broadly. Tencent has announced it would double its AI investments to more than 36 billion yuan ($5.2 billion) in 2026, according to Bloomberg and IndexBox.</p>\n\n<p>For developers and enterprises evaluating AI infrastructure costs, the V4-Pro's pricing—even at standard rates—represents a meaningful shift in what is economically viable. Access to a 1 million token context window at a fraction of Western competitor prices opens up use cases that were previously cost-prohibitive, particularly for smaller teams and organizations in price-sensitive markets.</p>\n\n<h2>Expert Reactions</h2>\n\n<p>Akshar Keremane, co-founder of Bangalore-based AI startup O-Health, described the practical significance of DeepSeek's approach for the developer community. "The pricing, open source availability and 1 million context window features all lower barriers for developers, startups and small enterprises," Keremane told Bloomberg. He added: "It allows users to experiment at a model capability and scale that wasn't available earlier."</p>\n\n<p>Ivan Su, senior equity analyst at Morningstar, highlighted what makes the V4-Pro launch strategically distinct from prior DeepSeek releases. "This is a framing that didn't exist with R1, and that alone tells you how much domestic competition has intensified," Su told CNBC.</p>\n\n<p>Haritha Khandabattu, analyst at Gartner, offered perspective on why the January 2025 R1 release had such outsized global impact—and implicitly, what would need to be true for V4 to replicate it. "January caused a broad, visible repricing because it changed global beliefs about frontier-model cost curves and China's competitiveness," Khandabattu told CNBC.</p>\n\n<p>Jensen Huang, CEO of Nvidia, offered a broader observation about the relationship between compute constraints and algorithmic innovation. Speaking on the Dwarkesh Podcast, Huang said: "The best AI researchers in the world, because they are limited in compute, also come up with extremely smart algorithms." The comment reflects a thesis that has become central to how observers interpret DeepSeek's trajectory: that operating under hardware constraints has pushed the company toward efficiency-driven breakthroughs rather than raw scale.</p>\n\n<h2>What's Next</h2>\n\n<p>The immediate milestone to watch is May 5, 2026, when DeepSeek's 75% promotional discount on V4-Pro API access is scheduled to expire. Whether the company extends the promotion, makes elements of it permanent, or allows pricing to revert to standard rates will signal how aggressively it intends to compete on cost going forward.</p>\n\n<p>The outcome of DeepSeek's first external funding round is also a significant variable. If Tencent, Alibaba, or other investors close a deal at a valuation above $20 billion, it would mark a substantial institutional endorsement of DeepSeek's model and strategic direction—and provide capital that could accelerate its roadmap. The reported reluctance to cede a 20% stake to any single investor suggests DeepSeek's leadership is thinking carefully about the terms of that capital, not just the amount.</p>\n\n<p>On the technical side, DeepSeek's own acknowledgment that V4-Pro trails GPT-5.4 and Gemini 3.1 Pro by three to six months sets an implicit benchmark. Closing that gap—or demonstrating that its cost advantages make it a viable alternative even at a capability discount—will be central to how the model is adopted by developers and enterprises over the months ahead.</p>\n\n<p>The broader AI price war DeepSeek has helped catalyze shows no signs of abating. For the companies that cannot compete on price and must compete on performance, differentiation, or ecosystem lock-in, the pressure will only intensify as open-weight models with million-token context windows become available at fractions of a cent per million tokens.</p>\n\n<p>For more tech news, visit our <a href=\"/news\">news section</a>.</p>\n\n<h2>What This Means for Your Health and Productivity</h2>\n\n<p>As AI tools become dramatically cheaper and more capable, the apps and platforms that power your daily health tracking, focus routines, and personal optimization stand to benefit directly—from smarter coaching to more personalized recommendations, all at lower cost. The race to the bottom on AI pricing is, for end users, a race toward better tools. At Moccet, we're building with that future in mind. <a href=\"/#waitlist\">Join the Moccet waitlist to stay ahead of the curve.</a></p>", "excerpt": "DeepSeek has cut API fees for its newly released V4-Pro model by 75%, valid until May 5, 2026, while permanently slashing input cache hit prices by 90% across all models. The moves have rattled rival Chinese AI firms and reignited the broad price war that began when DeepSeek's R1 model upended global AI markets in January 2025. V4-Pro, the largest open-weight model currently available, features 1.6 trillion parameters and a 1 million token context window—at pricing that dramatically undercuts Western competitors.", "keywords": ["DeepSeek V4-Pro", "AI price war", "DeepSeek API pricing", "Chinese AI models", "open-weight AI models"], "slug": "deepseek-v4-pro-75-percent-price-cut-ai-price-war" } ```