Money & Markets

DeepSeek Now Bills Search and Fetch Calls, Agent Tool Calls Included

By Willa Ravenwire · · 2 minUpdated

WIRE

Why it matters. The rate is small and clearly published, and failures cost nothing. The open question is volume, so check your usage logs before assuming built-in search is a rounding error.

If your coding agent runs on the DeepSeek API, its web lookups now show up on the bill.

The price. According to DeepSeek’s October 17 release notes, the Search and Fetch APIs opened to everyone at $0.75 per 1,000 calls for each endpoint. As Neomanex reports from DeepSeek’s Search and Fetch guide, each endpoint charges for successful requests only, rather than for tokens or result counts, at $0.75 per “1,000 successful requests” each.

What changed on November 1. The release notes say calls made through server-side tools in DeepSeek Harness, Claude Code and Codex were free for DeepSeek API users, with billing set to start October 31, 2026, at 16:00 UTC. Neomanex quotes the guide giving the same moment in Beijing time: charging for those calls begins at “00:00 Beijing time on November 1, 2026.”

What is not billed. Per the guide as Neomanex quotes it, a request counts only when it returns HTTP 200 with results. Failed requests, empty search results and timeouts are free.

Limits. Neomanex also lists separate rate limits from the guide: Search allows 100 queries per second and Fetch allows 25, both counted per account rather than per key. That shared Fetch cap matters if several agents run under one account.

What powers it. DeepSeek’s release notes say the service runs on its in-house search engine, which the company says serves hundreds of millions of requests a day worldwide.

The budgeting gap. Neomanex points out that DeepSeek gives no typical request count per agent session. That means the per-request price is clear, but the cost of a task depends on how many lookups your agent chooses to make. Its advice is to log those requests per session so you know your own volume.

Nothing changes in your code. Model calls work the same way. The difference is a second metered line next to token charges.

What’s real here: the rate is small and clearly published, and failures cost nothing. The open question is volume, so check your usage logs before assuming built-in search is a rounding error.

Share on X

Sources

By Willa Ravenwire · The AI Bandwagon

Next from the wirePope Leo XIV on AI: Machines Are Tools, Not Persons

More in Money & Markets

  1. Money & Markets · 4 min

    The $828 Billion AI Spending Bill Finally Rattled Wall Street. Wobble or Warning?

    Five tech giants plan to spend as much as $828 billion on AI infrastructure this year, and Friday's tech selloff showed investors are starting to flinch.…

    Nov 14, 2026

  2. Money & Markets · 2 min

    DeepSeek funding report puts round near 100 billion yuan

    The reports describe a substantial funding round, but their figures and closing status differ. Reuters' caveats belong beside its number. A later, more precise account deserves scrutiny rather than automatic…

    Nov 14, 2026

  3. Money & Markets · 2 min

    Samsung reports KRW 107.4 trillion quarterly operating profit

    Samsung's reported gains are concentrated in semiconductors, while its device division posted a loss. That split matters more than treating the group headline as a verdict on every business. Compare…

    Nov 13, 2026