# Sourcegraph Credit Pricing: Budget Deep Search and Code Finder Separately

Sourcegraph cut per-call credit prices and changed how service accounts are billed. Here's what to check before changing your team's budget.

By Paras Madan | 2026-11-20


On November 19, [Sourcegraph cut the credit price](https://sourcegraph.com/blog/credit-pricing-update) of Deep Search from 70 to 50 and Code Finder over MCP from 11 to 3. Smart Hovers no longer consume credits. The lower prices apply automatically, but there is a separate change for service accounts that run searches or other product workflows. A team can spend less per call and still see a higher total if its forecast left those accounts out.

## What a fixed credit budget buys now

| Sourcegraph feature | Previous credits per use | New credits per use |
| --- | ---: | ---: |
| Deep Search | 70 | 50 |
| Code Finder over MCP | 11 | 3 |
| Smart Hovers | 0.15 | 0 |

The percentages depend on what you are counting. Deep Search costs about 29% fewer credits per search, so a fixed budget buys 40% more searches. Code Finder costs about 73% fewer credits per call. As an example, 3,300 credits cover 300 Code Finder calls at the old rate or 1,100 at the new rate.

These are credit calculations, not dollar estimates. Sourcegraph's [pricing page](https://sourcegraph.com/pricing) describes pooled credits and annual contracts. Check your own allocation and agreement before translating any of these counts into spend.

## Use Code Finder when you know the repository

Suppose an engineer wants to find where your storefront validates a delivery address. If they know the exact symbol or configuration key, start with plain search. If they know the repository but not the implementation, [Code Finder](https://sourcegraph.com/docs/api/mcp#code-finder) can locate relevant sections and return file paths and line ranges. Pin the revision when comparing against a release rather than the default branch.

Deep Search serves a different question. If the engineer needs to trace which warehouse, address, and order services each apply their own rules, a file list is not yet an answer. [Deep Search](https://sourcegraph.com/docs/deep-search) can investigate the code and show the sources behind its findings. Ask it to establish one narrow fact per service and check those sources before using the result in a design document.

## Estimate credits per useful result

A single 50-credit Deep Search costs more credits than sixteen 3-credit Code Finder calls. That is not a reason to use sixteen calls by default: the engineer or coding assistant still has to assemble their findings, and requests to a separate coding model may carry separate costs.

Start with three recent questions from your team's backlog: one exact lookup, one location search within a repository, and one question that spans services. Note the number of calls, whether the citations answer the question, and how long the engineer spends joining the evidence. If a coding assistant repeats Code Finder calls with small wording changes, narrow the repository and question before increasing its allowance.

## Count service accounts as well as people

Sourcegraph says service accounts that use searches, code intelligence, or other product workflows now follow the same billing model as human users. Accounts used solely for platform administration remain excluded. Split your inventory by what accounts actually do. A maintenance account that also produces a nightly engineering report should not be treated as admin-only without checking.

In the Enterprise Portal, use the Credits page to inspect the balance and current rates. Deep Search documentation also points admins to Sourcegraph Analytics for a usage breakdown when that link is available. If the contract treatment of an account is unclear, take its actual workload and estimated volume to your account team rather than assigning it a made-up dollar rate.

## Keep the first rollout small

Start with one team and one existing read-only workflow. Record its weekly Code Finder and Deep Search totals along with the questions that had to be retried. Cheap calls make it easier to find more evidence, but repeating a poorly scoped search will still burn time. Increase the budget where the team gets a better answer, not where its tools make more requests.

I'm Paras Madan. I write about AI tools and how engineering teams use them.

If you need help explaining a technical product or a pricing change clearly, I can help with the writing.

[Book a 30-minute call](https://cal.com/paras-madan/30min).
