Agent Upstart

Cool things people are building with agents — and practical help getting the most out of them.

Playbooks · Coding tasks

Check your first DeepSeek search/fetch bill

Written November 2, 2026

Search and Fetch billing started November 1. Compare your first bill with the projection: only successful requests count, and 50 results cost the same as one.

What changed on November 1

Search and Fetch are billed per successful request, not by tokens and not by the number of results returned. DeepSeek's pricing page lists Search and Fetch each at $0.75 per 1,000 successful requests, and states that per-request billing for server-side calls through DeepSeek Harness, Claude Code, or Codex started at 00:00 Beijing time on November 1, 2026. The Search and Fetch guide states the counting rule: a request is billable only when it returns HTTP 200 and produces a result — Search must return a non-empty results array, and Fetch must return page content. Failed requests, empty search results, and timeouts are not billed. The count field controls how many results a Search returns, between 1 and 50, and does not change the price of the call. The counting and projection method is set out in Budget DeepSeek Search and Fetch calls; this page is only about comparing the first real bill with that projection.

Before you start

Have the projection ready: the successful Search calls and Fetch calls one representative task makes, and the projected cost for the period, counted by the method on the budget page. Write down the exact period the projection covers, and how many tasks ran in it. A bill for a different period, or a different number of tasks, cannot be compared with the projection until the periods and task counts line up.

Steps

  1. Re-count one task the same way as the projection. Run one representative task again and tally only successful calls, separately for Search and Fetch, exactly as the budget page describes. If the counts have changed since the projection — the agent searches more, fetches more, or retries more — the projection is stale, and the comparison in step 3 must use the new counts.
  2. Read what your DeepSeek account shows for the same period. Open your DeepSeek platform account and record the balance change and any usage breakdown it shows for the period, exactly as shown. The guide describes no report that counts Search and Fetch calls, so do not expect a call-count line. If the account shows only a balance total, record that, and record that no Search and Fetch split is shown — the tally from step 1 is then the record of what was called, and the balance total cannot be attributed to Search and Fetch alone.
  3. Compare the expected charge with what the account shows. Each successful call costs $0.00075. Multiply the successful Search and Fetch calls for the period by $0.00075: 1,000 successful Search calls and 1,000 successful Fetch calls come to $1.50. Compare that figure with any Search and Fetch amount the account breaks out. If the account shows no breakdown, compare the expected Search and Fetch figure with the tally only, and do not treat the whole balance change as the Search and Fetch bill — the same balance also pays for model token use, which is billed separately under model pricing.
  4. Investigate a mismatch before raising the cap. If the Search and Fetch amount shown is higher than the expected figure, check the usual causes in order: calls counted that failed, timed out, or returned empty results, which are not billed; a count setting mistaken for a price lever, which it is not; calls from another tool or harness using the same DeepSeek API key; and retries after rate-limit answers. Search is limited to 100 requests per second and Fetch to 25 per second at the account level, with a further per-domain limit on Fetch, and requests above a limit get HTTP 429 — retries after a 429 that then succeed are billed as successful calls. Fix the counting or the calls first; raise the cap only on a reconciled figure.
  5. Set the cap from the reconciled figure. Top the account up only to the cap amount the reconciled counts support, so the balance itself is the limit: when the balance runs out, the API answers 402, insufficient balance, and the calls stop. Fees are deducted from the topped-up balance or the granted balance, with the granted balance used first when both exist. Re-check after the next billed period and repeat the comparison; a projection that is never compared with a bill stays a guess.

What can go wrong

Limits

Related playbooks

Sources

Back to: Coding playbooks hub · Also: Budget DeepSeek Search and Fetch calls · DeepSeek, Exa, Brave, Tavily: which web-search API should your agent actually pay for now that search is metered · Work out what DeepSeek Search and Fetch calls will cost before billing starts on 1 November 2026