Perplexity is selling AI agents a cleaner way to read the web. Its new Context API takes a batch of URLs and returns each page as clean Markdown, raw HTML, a list of links or only the passages that answer a question, for $1 per 1,000 URLs. Perplexity’s blog post announcing it is dated 17 November 2026, and its developer account announced on X on 18 November that the API is live.
Quick Take
- Input: a list of URLs, up to 10 per request on the lowest usage tier, 15 on tiers 1 to 3 and 25 on tiers 4 and 5.
- Output: Markdown by default, plus HTML, links and query-matched highlights, with metadata such as title, author and published date.
- Price: $1 per 1,000 URLs, billed only for successful or not-modified results. Failed URLs are free.
- Freshness: by default it serves a stored copy fetched in the last two days; max_age_seconds set to 0 forces a live fetch.
- Perplexity’s own test: usable content for 92.3% of 1,000 URLs, 87.9% fact recall and 80.3% content precision.
In plain English
When an AI agent opens a web page, most of what it sees is clutter: menus, cookie banners, ads. That wastes tokens and confuses the model. The Context API strips that away and hands the agent just the article text, or just the paragraphs that matter for its question. Developers pay a tenth of a cent per page.
What happened
Perplexity Developers wrote on X: “Context API reads up to 10 URLs per call and returns clean .md, HTML, links, and snippets per page.” The launch follows the Monitor API Perplexity released on 16 November and extends its push to sell search and web infrastructure to developers, alongside products like Perplexity Computer’s multiagent mode.
How it works
- Endpoint: POST /v1/context with a list of urls and a formats field.
- Content extraction: Perplexity says a model scores each block of a page as main content or boilerplate and keeps only the content blocks.
- Highlights: passages closest to a query, in the page’s original wording, capped at 4,096 tokens per page with 1,024 as the default.
- Caching controls: cache_only never triggers a live fetch; if_modified_since lets a single-URL request return not_modified when a page has not changed.
- Per-URL status: a request returns HTTP 200 even when some pages fail, and each result carries its own status, such as ok, blocked, bot_challenge or crawl_timeout.
- SDKs: the Python SDK needs perplexityai 0.48.0 or later, the TypeScript SDK 0.43.0 or later.
Why it matters
For developers building agents, web reading is one of the most fragile parts of the stack. A managed endpoint with per-page statuses and flat pricing removes a lot of scraping code.
For costs, sending only relevant highlights to a model instead of whole pages can cut token spend on research and retrieval tasks.
For publishers, it is another commercial service built on fetching and storing their pages for AI use, which keeps questions about access and compensation open.
What’s still open
- The quality figures come from Perplexity’s own test, judged by a model it chose.
- How the API handles sites that block AI crawlers, beyond reporting them as blocked.
- How it compares on price and quality with existing scraping and reader APIs.
Sources
- Perplexity, “Context API extracts full context from any web page,” blog post, 17 November 2026
- Perplexity API documentation, Context API reference and quickstart
- Perplexity Developers on X, 18 November 2026
i10x Perspective
Perplexity’s core skill is turning messy web pages into something a model can use. Selling that skill directly to developers at a flat per-page price is a logical next step, and a sign that web access is becoming a standard building block for agent stacks. More from Perplexity and our Agents section.






