DeepSeek Cuts V4 Pro AI Model Prices by 75% in Aggressive Market Move

Chinese AI startup DeepSeek has launched its newest large language model with pricing that undercuts nearly every major competitor in the market. The Hangzhou-based company announced a 75% discount on its DeepSeek-V4-Pro model for developers, running through May 5, while simultaneously slashing input cache hit prices across its entire API lineup to one-tenth of the original cost.

The move represents a significant escalation in the ongoing price war between Chinese and American AI labs. According to reporting from Bloomberg, the discount structure is designed to aggressively capture developer mindshare during the model's preview phase. Input cache hits—where the model recognizes repeated or similar requests—now cost 90% less, which dramatically reduces expenses for applications that process recurring queries.

DeepSeek released two variants of the V4 model: the Pro version and a lighter Flash variant. The Pro model carries 1.6 trillion parameters and supports a context length of one million tokens. For perspective, that's roughly equivalent to processing 750,000 words in a single session. The Flash version, with 284 billion parameters, targets cost-sensitive applications that don't require maximum reasoning capability.

Pricing is where the real disruption occurs. DeepSeek's V4-Pro costs $3.48 per million tokens of output. Compare that to OpenAI's $30 or Anthropic's $25 for the same amount of work. Even fellow Chinese competitor Moonshot AI's Kimi model charges $4 per million tokens. The Flash variant drops to just $0.28 per million tokens. These numbers aren't theoretical—they're live API pricing that developers can access immediately.

Performance benchmarks place V4-Pro as trailing only Google's Gemini-Pro-3.1 in world-knowledge categories, according to the company's own documentation. The model reportedly beats other open-source alternatives in agentic coding and reasoning tasks. DeepSeek's technical report acknowledges the model falls marginally short of OpenAI's GPT-5.4 and Gemini 3.1 Pro, suggesting a developmental trajectory that trails state-of-the-art frontier models by approximately three to six months.

Hardware compatibility represents another strategic consideration. The V4 models have been adapted for Huawei's Ascend AI processors, which the Chinese tech giant has been developing to compete with Nvidia's GPUs. This matters because U.S. sanctions have restricted China's access to advanced American chips. Running on domestic hardware means Chinese developers can deploy these models without navigating export controls or supply chain uncertainties.

The timing of this release is notable. DeepSeek first captured global attention in December 2024 with its V3 model, which the company claimed it trained on just $5.6 million worth of processors. That announcement triggered a massive selloff in U.S. tech stocks as investors recalibrated assumptions about how much capital was actually required to build competitive AI systems. At one point, tech stocks lost $1 trillion in value following the news.

Now, more than a year later, the competitive landscape has shifted considerably. Alibaba, MiniMax, Knowledge Atlas, and Moonshot AI have all released high-performing open-source models this year. The market is no longer waiting for DeepSeek to define the category—it's already crowded. This release comes as the company reportedly seeks funding from Tencent and Alibaba in a round that would value the lab at $20 billion.

Industry analysts note this pricing strategy runs counter to broader sector trends. Both OpenAI and Anthropic have hiked prices and imposed rate limits to manage surging demand. Chinese developers have followed suit, increasing prices and removing unlimited usage subscriptions. DeepSeek's decision to slash costs instead suggests either confidence in scaling production costs downward or a deliberate strategy to establish market dominance before competitors can respond.

The company expects to lower V4-Pro prices further later in the year as Huawei scales up production of its new Ascend 950 AI processors. This projection assumes manufacturing ramp-up proceeds without significant delays—a reasonable assumption given Huawei's track record, but one that carries inherent risk.

For developers actually using these APIs, the physical experience matters. Lower prices mean more API calls, more iterations, more testing cycles. The 1 million token context length means you can upload entire codebases or lengthy documents without worrying about truncation. The cache hit discounts mean your production applications won't get hammered by repetitive queries from the same users. These aren't abstract benefits—they're measurable improvements in development velocity and operational costs.

Whether this pricing strategy is sustainable remains an open question. DeepSeek's parent company, High-Flyer, isn't short on cash, but the company needs to raise money to retain talent against competitors with larger valuations. The 75% discount ends May 5, and the long-term pricing structure after that remains unclear. Developers betting on these rates for production systems should plan for potential adjustments.

Whether users actually pay for it remains the real question.

Artūras Malašauskas is an AI Systems Integrator with 20+ years of production-grade web engineering experience. He has designed, shipped, and scaled enterprise Python/PHP systems for logistics, SaaS, and public-sector clients. For the past year, he has focused exclusively on AI integrations: deploying open-source LLMs, building generative media pipelines (image, audio, video), and engineering multi-agent workflows for real production environments. His standard: reproducibility, security, cost-efficient inference—no vaporware. He documents and evaluates emerging AI tooling, separating verified capabilities from marketing noise. Technical editor at: muza-ai.eu, ai-verslas.lt, ai-naujinos.lt Connect on LinkedIn