Pricing roundup: DeepSeek’s retrieval layer undercuts rivals by ~85%

DeepSeek has put a hard price on search and fetch for agents: $0.75 per 1,000 successful requests for each endpoint. That lands in a market where the usual price floor is $5 and up, which makes this look less like a tweak and more like a direct hit to everyone charging more for retrieval.

The company launched the Search and Fetch APIs on October 17, alongside DeepSeek-V4.1-Flash-1017. DeepSeek is calling the new model a Pareto improvement over the earlier Flash checkpoint, and it claims a 77.0 score on DeepSWE v1.1, ahead of Claude Opus 5 at 74.0. That model release may get the headlines. The pricing is the part that changes the math.

DeepSeek says failed requests, empty results, and timeouts won’t be billed. Per-request billing starts at 00:00 Beijing time on November 1, which is 16:00 UTC on October 31. Until then, calls through DeepSeek Harness, Claude Code, and Codex are free. The pricing is global, with no regional split, and each endpoint is capped at 5 requests per second per account.

The comparison is brutal. Brave Search API starts at $5 per 1,000 requests. Exa charges $7 for standard search and $12 for deep search. Tavily runs from $5 to $8 depending on plan. Perplexity Sonar starts at $5 and adds token costs on top. DeepSeek’s $0.75 is 6.7 times cheaper than the next cheapest provider and 9.3 times cheaper than Exa’s standard tier.

This is also the first time DeepSeek has charged for tooling beyond model tokens. That matters because the company already built its name on cheap inference, with Flash input tokens at $0.15 per million off-peak and cache hits as low as $0.003. Now it’s taking that same pressure campaign into retrieval, the layer agents use to ground answers in real data. In a market where infrastructure is getting commoditized, DeepSeek is trying to turn retrieval into another utility and make everyone else defend their margin.