DeepSeek’s new V4-Flash costs pennies compared to OpenAI and Anthropic models. Could ultra-low inference costs be changing the economics of enterprise AI?
DeepSeek is back doing the one thing that terrifies Silicon Valley CFOs: making Western AI look wildly overpriced.
The Chinese startup rolled out V4-Flash, its latest low-cost model, on Friday.
According to research firm Artificial Analysis, running a standardized test suite on V4-Flash costs about 3 cents. Running that same test on OpenAI’s GPT-5.6 Sol costs $1.86. On Anthropic’s Claude Fable 5, it costs $3.15. That is a hundredfold price collapse.
Now, let’s talk about the actual performance. V4-Flash won’t win awards for deep reasoning. It scored 50 out of 100 on Artificial Analysis’s Intelligence Index, placing it on par with Google’s Gemini 3.6 Flash and well behind frontier models. If you need an AI agent to architect complex software or solve heavy logic problems, you still pay the premium for top-tier models.
But most enterprise AI work isn’t complex logic. It is mundane plumbing: sorting support tickets, extracting data from invoices, or summarizing internal notes. A “50 out of 100” score works just fine for those routine tasks. When a model gets the job done, corporate buyers stop caring about benchmark bragging rights and start looking at their API bill.
That is where DeepSeek creates real market pressure.
The startup, reportedly prepping for an IPO, doesn’t need to beat OpenAI on raw intelligence. It just needs to bleed off the high-margin enterprise traffic that funds American AI labs. When everyday tasks cost virtually zero, convincing a CFO to pay $3 per test becomes a tough sell.
OpenAI and Anthropic built incredible reasoning engines, but DeepSeek just reminded everyone that in business, price per task usually wins.
