研究机构Artificial Analysis发布分析报告显示，中国AI创企DeepSeek于7月31日发表的新旗舰模型DeepSeek-V4-Flash具有显著的成本优势[1]。该模型的输入词元价格为每百万0.14美元，输出词元为每百万0.28美元，平均每次测试成本仅0.03美元[1]。相比之下，Claude Fable 5的测试成本为3.15美元，GPT-5.6 Sol为1.86美元，Kimi K3为0.86美元，DeepSeek-V4-Flash的成本比Claude Fable 5便宜逾100倍[1]。
性能方面，DeepSeek-V4-Flash在基准测试中获得50分，与Google的Gemini 3.6 Flash相当，但低于Claude Opus 5、GPT-5.6等高端模型[1]。这种成本与性能的平衡已在全球开发者中形成压倒性优势[2]。据OpenRouter平台最新周度数据，DeepSeek-V4-Flash成为该平台调用量最高的模型[2]。海外AI初创团队迁移至DeepSeek后，月度推理成本下降了60%至85%[2]。
DeepSeek已开始筹备首次公开募股（IPO）[1]。
Research firm Artificial Analysis has released an analysis showing that Chinese AI startup DeepSeek's newly launched flagship model DeepSeek-V4-Flash operates at substantially lower costs than rival systems [1]. The model, released on July 31st, carries an average test cost of just $0.03 per query—over 100 times cheaper than Claude Fable 5, which costs $3.15 [1]. Input token pricing for DeepSeek-V4-Flash stands at $0.14 per million tokens, with output tokens priced at $0.28 per million tokens [1]. By comparison, GPT-5.6 Sol costs $1.86 per test, while Kimi K3 runs at $0.86 [1].
Despite its cost advantages, DeepSeek-V4-Flash scores 50 points on the "Intelligence Index" benchmark—equivalent to Google's Gemini 3.6 Flash but below premium models like Claude Opus 5 and GPT-5.6 [1]. The model's extreme cost efficiency has already translated into market dominance; according to OpenRouter platform data, DeepSeek-V4-Flash has become the most-called model on the platform [2]. OpenCode CEO Jay reported a 30% single-day increase in usage following the model's launch [2], and overseas AI startups have reported monthly inference cost reductions of 60% to 85% after migrating to DeepSeek [2]. Comparable offerings from Silicon Valley firms like OpenAI, Claude, and Gemini are reportedly priced several times to nearly ten times higher than DeepSeek's [2]. DeepSeek is currently preparing for an initial public offering and plans to introduce a more powerful DeepSeek-V4-Pro version [1].
