Web search, distilled for your agents
One POST request returns clean, answer-ready web content. Mako feeds your LLM 20.2% fewer tokens than Tavily, measured on a public, replayable benchmark: same speed, better answers. Moby dives deeper when you need the whole story — and Krill, the free model, is unlimited (fair use).
Three models. Pick your depth.
Same request, one parameter.
Plankton · surfaceKrillThe free one
Built for trying things, and for projects that never bill.
The same engine, a smaller catch: the shortest set of sentences that still answers — ~700 tokens, leaner than anything else here. Fewer sources than Mako, and no search fee at all.
- Free & unlimited (fair use)
- ~700 tokens per search
- One request at a time
Surface · 0 mMakoThe fast one
Built for agents that search constantly.
Only the sentences that answer the question — 20.2% leaner than Tavily. Same answer, smaller bill.
- Answers in ~1–2 s
- ~1k tokens per search
- Best default for agents & RAG
Deep water · −6000 mMobyThe thorough one
Built for research that needs the whole story.
The full readable content of every source, with a detailed answer that cites them.
- Full page content per source
- Detailed, source-cited answers
- Best for deep research
"model": "mako" → "krill" or "moby"One request. Any language.
A single POST endpoint with two parameters: what to search and which model — Krill, Mako or Moby. Authentication is a bearer key. That’s the whole integration.
- Official SDKs: pip install serpdive, npm install serpdive
- Typed responses, typed errors, retries built in
- Extracted page content, not raw SERP noise
# pip install serpdive from serpdive import SerpDive client = SerpDive() # reads SERPDIVE_API_KEY response = client.search( "what happened between Trump and FIFA", model="mako", # "krill" → free & unlimited · "moby" → full page content ) print(response.results) # [SearchResult(url, title, date, content)]
Judged blind, against the live web
Same question to both APIs, payloads anonymized, an independent LLM picks which content better answers it. No cherry-picking possible because the judge doesn’t know who’s who.
- 01
Questions no model can memorize
Live questions authored by five third-party frontier models: breaking news, market moves, live sport, product availability… the things an agent has to search because it cannot already know them.
- 02
Blind A/B duels
Both APIs get the same question; their raw payloads are anonymized and the A/B order alternates. The judge never knows which provider produced which content.
- 03
An independent judge
A frontier LLM at temperature 0, anchored to a reference answer from a third search engine. Stale facts presented as current, wrong-language sources and hallucinated pages lose the duel.
- 04
Public, end to end
The full benchmark ships with the judge prompt, every question and every per-duel verdict publishedso you can rerun it yourself and check.
Mako vsMako vsTavily duel, July 2026 public run: 1,000 · GPT-5.4 · T 0. Mako wins 475 duels, Tavily 307, with 217 ties (60.7% of decided duels), while returning 20.2% fewer tokens (1,001 vs 1,255). Every prompt, question, raw payload and verdict is published; rerun the chain yourself and check.
Full benchmark on GitHubA search API bills you twice
Once for the search, again when every result flows through your LLM as input tokens. SERPdive undercuts both.
$75 on search fees + $19.05 on LLM input tokens.
Pay-as-you-go rates: Tavily $0.008 per basic search, SERPdive $0.005 per Mako search at the launch rate (list $0.007). Krill has no search fee on the free plan (fair use). Token math assumes ~1,255 tokens per Tavily response, ~1,001 for Mako and ~700 for Krill, and Claude Sonnet 5 input at $3/M tokens. Measured mean payloads; your queries will vary.
MakoKrillPay as you go, or commit and save. 25% off while we launch.
One rate, every feature — Krill stays free, unlimited (fair use), on every plan. Paid searches start with 1,000 free credits: no card, no plan to pick.
Launch −25% ends
For new builders
- Unlimited Krill searches (fair use)
- 1,000 API credits / month for Mako & Moby
- No credit card required
- Email support
Pay only for what you use
- First 1,000 credits every month are free
- No monthly minimum, cancel anytime
- Production rate limits (200 req/min)
- All three models: Krill, Mako & Moby
- Failed searches are never billed
4,000 credits / month · ≈ $0.0045 / credit
- Discounted rate: the more you commit, the less per credit
- Production rate limits (200 req/min)
- All three models: Krill, Mako & Moby
- Change or cancel anytime
- Priority email support
Frequently asked questions
The models, the pricing, the response format. Something we didn’t cover? Ask us directly.
Ready to dive in?
Give your agents better answers for fewer tokens. 1,000 free credits, no card, and 25% off every search while the launch clock runs.