NewsAPI.ai

What it is

A general news API with an academic pedigree, sold to finance among other markets. Two endpoints carry the product: getArticles searches around 150,000 outlets in more than 50 languages, and getEvents searches the same corpus clustered into events — a bombing, a merger, a rate decision — so one query returns the story rather than the forty write-ups of it. That clustering is the thing competitors here mostly do not have.

Articles arrive enriched: disambiguated concepts, categorisation, duplicate detection, article sentiment, social-share counts, authors, and the full body text. A separate text-analytics endpoint will run the same annotation, categorisation, sentiment and semantic-similarity models over text you supply.

Pricing

A slider on the pricing page, and the steps are public: free, then $90, $150, $220, $400, $700, $1,300 and $3,000 a month for 5K, 10K, 20K, 50K, 100K, 200K and 500K tokens, then custom. The displayed plan rotates on a timer, so what you see first depends on when the page finished loading — read the slider position, not the headline.

Tokens rather than requests is the unusual part, and the exchange rate is the pricing. One token buys a search of the last 30 days returning up to 100 articles. The same search against the archive costs 5 tokens per year covered, so a decade-wide query costs 50 and the $90 plan buys 180 of them. Event searches cost five times as much again. Paging is a fresh search and charged again. Nothing rolls over month to month; overage is $0.015 a token if you disable the block, and rejected requests if you do not.

Data & coverage

Over 150,000 news outlets and 50-plus languages, with a claimed 99.99% uptime SLA, CORS enabled for all origins and real-time article availability. The archive starts in 2014, which is the oldest date on the pricing page and the only one the vendor states. Free accounts see the last 30 days and nothing older.

What you do not get is anything financial in the schema. There is no ticker, no ISIN, no exchange and no instrument of any kind — entities are Wikipedia concepts and you filter on their URIs. For a portfolio-shaped query that is a mapping layer you build and maintain.

Integrations

REST, plus a first-party Python SDK (eventregistry, by Gregor Leban of Event Registry, 10.0 in September 2026) and a Node client. The SDK is the documented path and wraps paging as an iterator. A sandbox on the site builds a query in the browser and hands back the code for it. There is also an official MCP server — newsapi-mcp on npm, run with npx, from the vendor's own GitHub organisation and at 1.3.1 in March 2026 — which searches articles, tracks events and pulls topic pages from a prompt. What there is not is a websocket or a webhook push: everything else here is polling.

Limitations

• No ticker or security identifiers at all, which is the single fact that decides whether this belongs in a market-data stack.
• The archive is charged by the year you reach into. Depth is available and expensive on purpose; a backfill is a budget question, not a plan question.
• The free tier is a demo, not a tier. 2,000 tokens once, 30 days of content, and terms that forbid any use producing commercial or institutional value even indirectly.
• Redistribution is barred — no sharing, selling, sublicensing or embedding of third-party content without your own licence from the publisher, and no building a competing aggregation or event-detection service. Terms last updated 5 November 2025, Slovenian law, disputes in Slovenia.
• Sentiment is per article, not per entity. A piece comparing two companies gets one score.

Alternatives

Marketaux scores sentiment per ticker rather than per article, on a free tier and then from $29 a month; Finlight resolves companies to tickers, ISINs and OpenFIGI identifiers. Both are narrower corpora. If the event clustering is what you came for, nothing else in this catalogue does it — GDELT is the free approximation, with its own archive back to 2015 and no support at all.

Compare it against the rest of the market data APIs.