Jina
A URL-to-markdown Reader that benchmarks at the top of its class, plus search that hands back full-page markdown in a single call.
Where Jina lands
Composite scores out of 10, from benchmark runs where every provider faces the identical task corpus. Rank is within that capability's field; the newest run backing this page is 2026-06-22. Full method: how we measure.
- Text accuracy
- 0.78
- Latency p50
- 263 ms
- Cost
- $0 / useful
- Errors
- 0%
- Recall@10
- 0.60
- Latency p50
- 1.4 s
- Cost
- $0.00033 / useful result
- Errors
- 0%
- Correctness
- 1.00
- Citation faithfulness
- 0.11
- Latency p50
- 2.6 s
- Cost
- $0.02 / call
- Errors
- 0%
The honest pitch
Reach for it if a URL needs to become clean, LLM-ready markdown, or you want search and read combined in one call.
Skip it when high search RPM or lots of results per query matter — the default is ≤5 results and RPM runs low.
About Jina
Jina concentrates on converting the web’s live mess into input a model can trust. Point the Reader at any URL and pristine markdown comes back with boilerplate already stripped; the search endpoint goes further, executing a query and handing back full page content for every result in one call. For agents and RAG pipelines that need dependable text instead of raw HTML, it ranks among the strongest tools available, and its own embedding and reranking models complete the retrieval stack end to end.
Reaching it through NativePort
Identical paths, parameters and responses to Jina's own documentation — the gateway holds the upstream credential and meters your balance at Jina's real published usage price.
curl https://api.nativeport.ai/jina/<native-path> \ -H "Authorization: Bearer $NATIVEPORT_API_KEY"