Case study · measured 2026-08-29
On 29 August 2026 we ran identical prompts through both engines, using our own paid API key on each. We wrote down everything: verdicts, page counts, queue times and real cost. Nothing here is estimated. The raw data is downloadable at the bottom, so you can check our arithmetic or rerun it yourself. Where they win, we say so.
Gumshoe
page read + LLM synthesis (gpt-oss-120b). Measured cost per URL read: $0.0016.
4/5
SLAYGENT · SLAYGENT.CO
cached multi-page read + light classifier, site-anchored. Public list price per URL: $0.0045.
Part 1 · qualification, identical prompt
Identical prompt (removal test) sent to both engines. Expected verdicts set by hand before running. One pure-play agent vendor, three products that added agent features on top, one with none.
| URL | EXPECTED | GUMSHOE | SECONDS | SLAYGENT | QUEUE |
|---|---|---|---|---|---|
| sierra.ai | core | core OK | 8.5s | core OK | 5s |
| attio.com | not-core | exclude OK | 7.2s | watchlist OK | 19s |
| notion.so | not-core | watchlist OK | 17.6s | watchlist OK | 15s |
| clay.com | not-core | watchlist OK | 7.9s | watchlist OK | 14s |
| basecamp.com | exclude | exclude OK | 6.2s | watchlist X | 15s |
BOTH ENGINES HANDLE QUALIFICATION WELL, AND SLAYGENT READS UP TO SIX PAGES PER CREDIT. THE ONE MISS: A GENERIC SAAS CALLED WATCHLIST INSTEAD OF EXCLUDED. THIS TEST IS TABLE STAKES FOR BOTH OF US.
Part 2 · capability, where the gap opens
GUMSHOE · PASSED
13 browser steps, 5 companies extracted, screenshots at every step; 12 screenshot exhibits returned with the report
55s · 42 credit(s)
SLAYGENT · FAILED
Returned page title only. Stated the content contains no directory listings.
GUMSHOE · PASSED
$950M Series E, May 2026, investor list, 4 sources
19s · 15 credit(s)
SLAYGENT · FAILED
Answered: no information about any recent funding. Anchored to the company domain even with web search enabled.
GUMSHOE · PASSED
Exact role titles with locations
7s · 1 credit(s)
SLAYGENT · PASSED
Exact role titles
GUMSHOE · PASSED
Returned exclude with reason: site unreachable, connection refused. Flagged similarly named companies as different entities.
1 credit(s)
SLAYGENT · NOT TESTED
not tested
Part 3 · what each action actually costs us
| Read one URL | $0.0016 | 1 credit |
| Browser fieldwork, per URL | $0.072 | 42 credits |
| Deep research dossier | $0.0228 | 15 credits |
one URL on a JS directory, real browser, 12 steps at $0.006 per step, Browser Use v2 measured 30 August 2026. 4 pages read + 2 outside searches + long synthesis, 19 seconds. We publish our costs because the pricing follows from them: 7,500 credits for $29, and you always know which of the three a case used.
WHERE SLAYGENT WINS
Fast, cheap qualification from a company site. Reads up to 6 pages per credit. Passed 4 of 5 qualification cases with our prompt. Their free tier is generous and their list price per URL is lower than ours on paper. If your whole workload is reading company sites and speed of setup matters more than reaching past the homepage, they are a reasonable pick and we will not pretend otherwise.
METHOD AND LIMITS
All figures measured by us on 29 August 2026 using both vendors public APIs with our own keys. Competitor behaviour may change. Raw responses available on request.
Slaygent public API v1/research, our own paid API key, 29 August 2026
1,000 free credits a month, no card. Run the same five URLs and compare what comes back.