DeepSeek cuts the bill for AI agents and forces OpenAI to defend the premium

DeepSeek has not introduced a new laboratory stunt. It has sent a more uncomfortable signal to rivals: a price that buyers can build into a budget. According to The Decoder, the company is making the 75 percent discount on its top V4-Pro model permanent. The new standing price is $0.435 per million input tokens.

That figure matters only when it is placed beside the competition. The Decoder reports that DeepSeek V4-Pro is at least 11.5 times cheaper than GPT-5.5 on input tokens and more than 34 times cheaper on output tokens. In practice, output is often where the bill starts expanding: the model writes plans, responses, checks, retries, and tool-facing instructions, not just a neat final answer.

That is why the move matters beyond ordinary chatbots. For a simple user question, a lower model price may be useful without being decisive. For agentic systems, where one request can trigger dozens or hundreds of model steps, tokens become an operating cost, almost like compute infrastructure. In that context, the official DeepSeek API documentation and public OpenAI API pricing pages are useful reference points, even though the specific comparison here comes from The Decoder’s reporting.

The important word is permanent. A discount can be dismissed as acquisition strategy: pull developers in, fill usage pipelines, then change the terms later. A standing price sends a harder message. DeepSeek wants developers, startups, and larger buyers to treat lower cost as a planning assumption, not a temporary window.

That puts Western model providers in a tighter position. If a product does not need the most expensive model available, but does need a capable model with very low output costs, the service economics change. Customer support systems, internal analytics agents, document summarization, code generation, and multi-step automation can become cheaper to run or easier to scale aggressively without the same fear of the token bill.

Of course, the market is not decided by token price alone. Buyers still evaluate answer quality, latency, API reliability, model availability, legal terms, data privacy, and integration with existing tools. For many Western organizations, security, procurement, and regulatory requirements do not disappear because the meter is cheaper. The broader issue also connects to how AI services are embedded inside production systems, from agent workflows to internal tools that must remain predictable, measurable, and available.

Still, DeepSeek’s move changes the tone of the conversation. The AI race is no longer only about who has the stronger demo or the higher benchmark score. It is increasingly about the price per unit of useful intelligence. If DeepSeek can sustain this gap, higher-priced models will need to explain much more clearly what users receive in exchange for a multiple of the cost.