Otto vs Jina Content Extraction

This page documents a practical comparison between two extraction approaches:

• Otto — browser DOM extraction from a live Chrome session via otto extract-content
• Jina — remote page fetch and parse via https://r.jina.ai/<url>

The comparison uses three representative URLs and captures both raw outputs and performance metrics.

Methodology

For each URL we ran:

• otto extract-content <url> --json
• curl -s "https://r.jina.ai/<url>" -w '\nTIME_TOTAL:%{time_total}\n'

This test deliberately omitted an API key, using the anonymous r.jina.ai endpoint. Jina response times can be cached, so only the first request duration is used in this comparison.

Raw output files were saved under docs/guides/outputs/:

• otto_<slug>.json
• jina_<slug>.json

This comparison focuses on:

• extraction content quality
• UI/noise removal
• comment/thread capture
• request duration
• site-specific coverage and blocking behavior

Jina rate limit context

Jina enforces rate limits in two dimensions:

• RPM — requests per minute
• TPM — tokens per minute

Limits are enforced per IP or per API key, whichever threshold is reached first. Anonymous requests are tracked by IP; authenticated requests are tracked by key.

The dashboard lists endpoint-specific limits such as:

• Reader API (https://r.jina.ai): URL-to-LLM-friendly-text extraction with anonymous and authenticated rate limits.
• Search API (https://s.jina.ai): web search + parsing with fixed token costs.
• Embedding API (https://api.jina.ai/v1/embeddings): both RPM and TPM apply, with token usage based on input size.
• Reranker API (https://api.jina.ai/v1/rerank): also RPM + TPM.
• Classifier APIs (/v1/train, /v1/classify): request and token budgets apply with separate few-shot and zero-shot limits.
• Segmenter API (/v1/segment): request-limited and token usage not counted.
• DeepSearch (https://deepsearch.jina.ai/v1/chat/completions): conversational search with a separate RPM budget.

These limits are relevant because Jina is a hosted remote service, while Otto performs extraction in the user's own browser.

1. LinkedIn post

URL: https://www.linkedin.com/posts/techstars_ny-tech-week-were-coming-for-you-whether-activity-7454971517832011776-fXua

Otto result

• durationMs: 3009
• contentLength: 7821
• Output: cleaned markdown extraction of the post content and visible page elements.
• Notes: Otto produced DOM-based markdown from the live browser page.

Jina result

• time_total: 4.638154
• Output: a mixed guest-page result labeled as markdown, but still containing LinkedIn login flow and UI shell markup.
• Notes: even without an API key, Jina returned a LinkedIn guest-page wrapper rather than a clean, distilled post summary.

Comparison

Aspect	Otto	Jina
Request duration	3.01 s	4.64 s
Content length	7,821 chars	Mixed guest-page markdown
Output type	Browser DOM markdown	Raw LinkedIn guest-page content
Comments	N/A	N/A
UI noise	Minimal	High (login/signup/navigation/footer)
Best fit	agent-ready content extraction	raw profile/guest page fetch

2. Reddit post

URL: https://www.reddit.com/r/LocalLLaMA/comments/1t1lfhj/minimax_m27_awq4bit_on_2x_spark_vs_2x_rtx_6000/

Otto result

• durationMs: 2313
• contentLength: 27791
• Output: markdown extraction of the Reddit thread, including post content and comment structure.
• Notes: Otto was able to extract the live page DOM from the target Reddit thread.

Jina result

• time_total: 1.815514
• Output: a Jina markdown error page showing that Reddit returned 403 Forbidden and the request was blocked.
• Notes: Jina could not retrieve the Reddit thread content anonymously.

Comparison

Aspect	Otto	Jina
Request duration	2.31 s	1.82 s
Content length	27,791 chars	Blocked error page
Output type	Markdown extraction	Blocked/restricted page notice
Comments	Extracted post comments	No usable content
UI noise	Low	High (block notice)
Best fit	browser DOM extraction for community content	not reliable for anonymous Reddit scraping

3. Blog post

URL: https://dennishodgson.blogspot.com/2025/05/photographic-highlights-202425.html

Otto result

• durationMs: 3017
• contentLength: 39427
• Output: cleaned markdown extraction of the blog article content.
• Notes: Otto removed page chrome and extracted the article from the live DOM.

Jina result

• time_total: 5.977519
• Output: actual markdown content for the article, including text and image links.
• Notes: Jina succeeded on this publication page and returned a markdown-ready article.

Comparison

Aspect	Otto	Jina
Request duration	3.02 s	5.98 s
Content length	39,427 chars	Markdown article content
Output type	Browser DOM markdown	Markdown-ready page extract
Comments	N/A	N/A
UI noise	Minimal	Low/medium (still includes some extracted navigation text)
Best fit	article extraction from live browser state	raw public blog extraction

Overall conclusions

• Otto is stronger for agent-ready extraction when the page is already loaded in a browser. It consistently delivers cleaned markdown from the live DOM.
• Jina can work for public blog pages, but it is less reliable for sites with access controls or anti-scraping protection.
• Reddit in particular failed anonymously through Jina with a 403 Forbidden block page.
• LinkedIn via Jina still returned guest-page wrapper content, not a distilled post payload.
• Otto’s extraction times were 2.3–3.0 seconds on this set, while Jina’s first-request times ranged from 1.8 to 6.0 seconds.
• Only the initial Jina request timing is meaningful here. Jina caches results, so repeated calls can appear artificially fast.

Evidence files

Raw capture files are available in docs/guides/outputs/:

• otto_posts_techstars_ny-tech-week-were-coming-for-you-whether-activity-7454971517832011776-fXua.json
• jina_posts_techstars_ny-tech-week-were-coming-for-you-whether-activity-7454971517832011776-fXua.json
• otto_r_LocalLLaMA_comments_1t1lfhj_minimax_m27_awq4bit_on_2x_spark_vs_2x_rtx_6000.json
• jina_r_LocalLLaMA_comments_1t1lfhj_minimax_m27_awq4bit_on_2x_spark_vs_2x_rtx_6000.json
• otto_2025_05_photographic-highlights-202425.html.json
• jina_2025_05_photographic-highlights-202425.html.json

These files preserve the exact initial tool outputs, timing measurements, and any site-specific noise observed during the comparison.