DeepSeek launched a search API that costs $0.75 per thousand successful requests. That works out to less than a tenth of a cent per query.

Compare that to Perplexity’s search API, which runs about $5 per thousand requests for their standard tier. DeepSeek just undercut them by 85%.

The pricing follows DeepSeek’s pattern. Their R1 reasoning model costs about $0.55 per million input tokens and $2.19 per million output tokens. GPT-4 runs closer to $2.50 and $10 per million tokens. DeepSeek keeps showing up with models that perform respectably at prices that make OpenAI’s look inflated.

What you actually get

DeepSeek’s search feature does what you’d expect: it takes a query, searches the web, fetches content from relevant pages, and returns it in a structured format your code can use. The API documentation shows you can integrate web search directly into your LLM applications without building your own crawler or paying for a traditional search API.

The $0.75 per thousand pricing only covers successful requests. Failed requests don’t count against your bill, which is cleaner than paying for queries that error out.

According to their API docs, you can use this to build RAG systems that pull in current information, fact-check LLM outputs against live web data, or just add search to a chatbot without the OpenAI or Perplexity premium.

The catch

There’s always a catch with this kind of pricing. DeepSeek is a Chinese company, which means your data flows through Chinese infrastructure. That matters if you’re building anything that handles user data subject to GDPR, HIPAA, or similar regulations. It matters even more if you work in defense, government, or any industry where data residency is a hard requirement.

The other catch is reliability. DeepSeek’s models are good, but their API uptime and support don’t have the track record of OpenAI or Anthropic. If your application breaks at 2am, you’re probably not getting the same level of support you’d get from a U.S. company with enterprise SLAs.

Why it matters

This kind of pricing pressure forces everyone else to compete. When DeepSeek drops search to $0.75 per thousand, Perplexity has to justify why their API costs nearly seven times more. Maybe they can, with better results or reliability. But they have to make that case now.

For developers, it means search-augmented LLMs just got cheaper to experiment with. A thousand queries for 75 cents means you can prototype something, run it through real usage, and see if it works before you’ve spent enough to buy a coffee. That changes what’s worth trying.

The broader pattern here is that AI pricing keeps falling faster than most people expected. DeepSeek keeps accelerating that trend. If you’re building something that depends on LLM costs staying high, or you’re betting that only premium providers can deliver quality results, this is the trend that breaks your assumptions.