Research jobs that cite their sources
Describe a question. AlterLab plans the searches and fetches the pages. Your own model key writes the synthesis, so you control the model and its cost.
Up to 5,000 free requests
We fetch, you reason
Plan
Describe the question and pick an effort tier. AlterLab plans the searches.
Fetch
Searches run and each source page is fetched at the lowest successful tier.
Reason
Your own model key writes the synthesis from the cited pages. Model tokens are never billed by AlterLab.
Bring your own model key
AlterLab bills the searches and the pages it fetches. The reasoning runs on your model provider account, so model tokens are never billed by AlterLab.
The scoreboard, kept in the open
A reproducible harness compares providers on answer quality, content coverage, latency and cost per correct answer. Numbers appear here only from dated runs.
| Set | Metric | Status |
|---|---|---|
| Protected-answer set | Answer-in-content rate | Pending scoreboard |
| Freshness set | Recall of last-7-day events | Pending scoreboard |
| Public QA subsets | Recall@10 of gold URL | Pending scoreboard |
| All sets | p50 / p95 latency, cost per correct answer | Pending scoreboard |