中国开源大模型通过算法架构创新实现极低成本，正在全球大模型市场引发一场价格战。Nous Research宣布DeepSeek V4 Flash在其平台启动一折促销持续7天[1]，调用成本比Fable 5便宜千倍以上，正价也维持百倍左右的优势[1]。8月1日，DeepSeek V4 Flash消耗8万亿token，其中5万亿为免费额度、3万亿为付费额度[1]。
DeepSeek、Qwen等国内创业公司利用稀疏激活、注意力机制重构等基础研究创新实现了价格优势[1]，正在冲击传统大厂的商业模式。大厂CEO表示当前任务是以超过行业平均的速度获取市场份额，"利润率是我们的第二目标"[1]。这一技术突破不仅引发行业价格竞争加剧，还带来了流量机制变革、人才培养危机等深层连锁反应[1]，如同核时代到来时那样，蕴含着难以预知的深远社会影响[1]。
Chinese open-source large language models are triggering a global pricing collapse through algorithmic innovations, forcing established players to reassess their commercial strategies. DeepSeek V4 Flash and similar models have achieved dramatically reduced operational costs, enabling aggressive price competition that disrupts traditional business models across the sector.
Nous Research launched a seven-day promotional campaign offering DeepSeek V4 Flash at a 90% discount [1]. The cost advantage extends far beyond promotional pricing: DeepSeek V4 Flash's standard rates undercut competitor Fable 5 by over a hundredfold [1]. According to OpenCode, on August 1st, DeepSeek V4 Flash consumed 8 trillion tokens, comprising 5 trillion free tokens and 3 trillion paid tokens [1].
The competitive pressure reflects fundamental technological breakthroughs rather than unsustainable discounting. Chinese open-source teams have leveraged sparse activation and attention mechanism restructuring innovations to achieve these cost efficiencies [1]. Industry leaders have acknowledged the shift, with major tech company executives stating that capturing market share at above-average industry speeds is the current priority, while positioning "profit margins as our second objective" [1]. The rapid technological advancement carrying unpredictable deep societal consequences has prompted observers to invoke the historical Oppenheimer metaphor, suggesting this transformation marks a pivotal inflection point comparable to the nuclear age [1].
