Analysis
DeepSeek has formally released its V4 Pro model, priced at $1.32 per million input tokens and $3.96 per million output tokens -- roughly nine and fourteen times the price of its own V4 Flash tier -- as the Chinese AI lab steps up hiring, compute spending and fundraising, according to Reuters. The release follows DeepSeek's first outside financing round, roughly $7.4 billion closed in June, and comes as the company is reportedly in talks for a new round that would value it near $74 billion, a dramatic reversal for a company that had avoided external capital entirely until this year.
DeepSeek's own history makes the pricing shift notable. The company's R1 model, released in January 2025, triggered a global sell-off in AI and chip stocks by demonstrating frontier-level reasoning performance at a training cost far below what OpenAI, Anthropic and Google were reportedly spending -- a claim that rattled the market's assumption that only companies with billions of dollars in dedicated compute budgets could compete at the frontier. V4 Pro's price hike suggests DeepSeek is now willing to trade some of that cost-disruption positioning for margin, betting its agentic capability gains justify charging closer to what Western labs charge rather than consistently undercutting them.
“The $74 billion prospective valuation is the number worth scrutinizing hardest.”
The competitive field V4 Pro enters is the most crowded it has ever been: OpenAI's GPT-5.6 Sol, Anthropic's Claude models, Google's Gemini 3.1 and Alibaba's Qwen 3.8 are all shipping agentic-capability updates within weeks of each other, and DeepSeek's claim to differentiation -- reasoning performance at a lower price -- gets harder to sustain the closer its own pricing moves toward the field it built its reputation undercutting. Domestically, DeepSeek also faces faster-moving rivals like Alibaba's Qwen team and Moonshot AI's Kimi models, both of which have narrowed the capability gap that gave DeepSeek its initial breakout moment.
The $74 billion prospective valuation is the number worth scrutinizing hardest. DeepSeek has never disclosed audited revenue, operates under Chinese data and export-control constraints that limit its access to the newest Nvidia chips, and its path to a valuation anywhere near OpenAI's or Anthropic's rests almost entirely on continued model-quality parity rather than a demonstrated enterprise revenue base comparable to its U.S. peers. A company moving from zero outside capital to a $74 billion target valuation within months is pricing in geopolitical and strategic value -- China's clearest frontier-AI champion -- as much as product economics. Founder Liang Wenfeng, who built DeepSeek out of the quant hedge fund High-Flyer, has said little publicly about the fundraising plans, keeping the company's usual low-key posture even as its valuation ambitions have gone from nonexistent to a figure comparable with OpenAI and Anthropic's own early-2025 marks within a single year.
The real test is whether enterprise customers outside China actually adopt V4 Pro at its new price point, or whether DeepSeek's real leverage remains open-weight releases that get adopted indirectly through fine-tunes and derivatives rather than direct API revenue -- the model by which most of DeepSeek's global usage has spread until now.
