--- source_url: "https://qveris.ai/guides/financial-news-api-for-ai-agents/?lang=en" title: "Financial News API Benchmark: 7 Providers Tested" mirrored_at: 2026-08-09T01:38:51.264Z host: qveris.ai cited_in_42a: true mirror_canonical: "https://index.42a.ai/qveris.ai/guides/financial-news-api-for-ai-agents/index__q__lang_en" --- > **Original source:** https://qveris.ai/guides/financial-news-api-for-ai-agents/?lang=en Live Financial News API Benchmark金融新闻 API 实测基准 A dated, reproducible comparison of seven suppliers across company discovery, China/HK and specialist news, macro and cross-asset retrieval, strict ticker lookup, language coverage, relevance, freshness, fields, and latency.一份带日期、可复现的七家供应商实测报告，覆盖全球公司发现、中国/香港与专业新闻、宏观和跨资产检索、严格 Ticker 查询、语言覆盖、相关性、新鲜度、字段完整度与延迟。 By QVeris Research · Live tests run August 3, 2026 · Updated August 3, 2026QVeris Research · 实测日期：2026 年 8 月 3 日 · 更新日期：2026 年 8 月 3 日 Contents目录 - [What we tested测试范围](#what-is) - [Evaluation standard评价标准](#selection-framework) - [Best observed fits实测最佳适配](#compared) - [Ticker and language resultsTicker 与语言结果](#use-cases) - [Selection guidance选择建议](#unified) - [FAQ常见问题](#faq) TL;DR There is no honest universal winner. In this dated comparison, **Finlight** was the strongest structured ticker-news option, with 24/24 processable responses and relevant results in both rounds for all nine listing markets. **Linkup** produced the broadest observed requested-language retrieval, but it is a Web retrieval product rather than a strict financial news feed. **Gildata** and **Caidazi** were strongest in the tested Chinese specialist workflows. Keep product shapes separate: global discovery, strict ticker filtering, multilingual retrieval, sentiment fields, and regional specialist feeds are different buying decisions. 不存在诚实可信的“全能冠军”。在本次带日期的实测中，**Finlight** 是最强的结构化 Ticker 新闻方案：24/24 次响应可处理，九个上市市场两轮均返回相关结果。**Linkup** 的请求语言检索范围最广，但它属于 Web 检索产品，并非严格的金融新闻 Feed。**Gildata** 与 **Caidazi** 在测试的中文专业新闻工作流中表现最好。 选型时必须区分全球发现、严格 Ticker 筛选、多语言检索、情绪字段与区域专业 Feed，不能把它们压成一个通用评分。 **Provider profiles:** [Brave Search](https://qveris.ai/providers/brave_search), [Caidazi](https://qveris.ai/providers/caidazi), [EODHD](https://qveris.ai/providers/eodhd), [Finlight](https://qveris.ai/providers/finlight.me), [Gildata](https://qveris.ai/providers/mcp_gildata), [Linkup](https://qveris.ai/providers/linkup), and [NewsAPI](https://qveris.ai/providers/newsapi). **供应商页面：**[Brave Search](https://qveris.ai/providers/brave_search)、[Caidazi](https://qveris.ai/providers/caidazi)、[EODHD](https://qveris.ai/providers/eodhd)、[Finlight](https://qveris.ai/providers/finlight.me)、[Gildata](https://qveris.ai/providers/mcp_gildata)、[Linkup](https://qveris.ai/providers/linkup) 与 [NewsAPI](https://qveris.ai/providers/newsapi)。 ## What we tested The broad comparison contains **316 completed live observations** across 158 tool-case cells, with every cell run twice. Cases cover companies, macro, crypto, forex, invalid input, and relevant specialist feeds across nine market regions and seven requested languages. The strict ticker comparison contains **44 completed observations** across 22 product-case cells: nine listing markets, one invalid ticker control, and provider-specific ticker-dialect controls. A market passed only when both rounds returned news relevant to the target company. The fixed public shortlist is Brave, Caidazi, EODHD, Finlight, Gildata, Linkup, and NewsAPI. A supplier remains visible even when a later retest falls below the gate, preserving comparability between editions. **Important:** a processable response, the expected empty/non-empty shape, and relevant news are separate metrics. Combining them would reward irrelevant fallback content and punish honest empty results. Metric Public calculation and pass rule Processable response rate Completed observations that returned a structurally usable response divided by all completed observations. This is a transport-and-shape measure, not proof that the returned news is relevant. Empty results still need the separate expected-shape and relevance gates. Expected-response rate Positive cases pass when at least one article is returned. The invalid-input control passes only when no article is returned. Relevant-result round rate Applicable live rounds returning at least one relevant top-five row divided by completed applicable rounds. Top-5 relevance Relevant rows among the first five results. A row is relevant when the requested entity or topic alias appears in its title, summary, or structured entity fields. Fresh-record rate Returned rows with a usable publication time inside the case's seven-day window divided by all returned rows; a missing time does not receive freshness credit. Core-field completeness Present values for title, URL, publication time, source, and summary divided by the expected core-field values in returned rows. Observable source diversity Distinct normalized source values per non-empty observation. Zero means no usable normalized source field, not zero upstream publishers. Duplicate rate Repeated normalized URLs or title-and-date identities divided by returned rows. Sentiment coverage Relevant rows with an exposed sentiment value divided by relevant rows. This measures availability, not sentiment accuracy. Requested-language rate Rows matching the requested language divided by rows whose language is explicit or detectable. The observation denominator is always shown. Requested-country rate Rows matching the requested country divided by rows with observable country metadata in cases requesting a concrete country. The observation denominator is always shown. Median latency Median end-to-end elapsed time across completed live observations. It is an observation from this run, not an SLA. Strict ticker-market pass A real market passes only when both rounds return news relevant to the target company through the provider's dedicated ticker or symbol input. Negative-control pass The invalid query or ticker returns no article. An irrelevant fallback is a failure, not a successful result. ## 我们测试了什么 公开的广义对比包含 **316 次已完成实时观察**，覆盖 158 个工具—案例单元，每个单元运行两轮。案例包括公司、宏观、加密资产、外汇、无效输入，以及九个市场区域和七种请求语言下的专业 Feed。 严格 Ticker 对比包含 **44 次已完成观察**，覆盖 22 个产品—案例单元：九个上市市场、一个无效 Ticker 对照，以及供应商专属的 Ticker 方言对照。只有两轮都返回目标公司相关新闻，市场才算通过。 固定公开候选名单为 Brave、Caidazi、EODHD、Finlight、Gildata、Linkup 与 NewsAPI。后续复测即使未达标也保留该行，以保证不同版本可比较。 **注意：**响应可处理、返回符合预期的空/非空形状、以及新闻相关，是三个不同指标，不能合并成一个成功率。 指标 公开计算方法与通过规则 可处理响应率 结构可用的已完成观察数除以全部已完成观察数。这只衡量传输与响应形状，不代表新闻相关；空结果仍需单独通过预期形状与相关性门槛。 预期响应率 正向案例至少返回一篇文章才通过；无效输入对照只有在不返回文章时才通过。 相关结果轮次率 至少返回一条 Top-5 相关结果的适用实时轮次，除以全部已完成适用轮次。 Top-5 相关度 前五条结果中的相关记录占比；请求实体或主题别名出现在标题、摘要或结构化实体字段中时，记录才算相关。 新鲜记录率 案例七天窗口内具有可用发布时间的返回记录占全部返回记录的比例；缺少时间的记录不计新鲜度。 核心字段完整度 返回记录中标题、URL、发布时间、来源和摘要的有效值，占预期核心字段值的比例。 可观察来源多样性 每个非空观察中的不同标准化来源数。零表示没有可用的标准化来源字段，不代表上游没有发布者。 重复率 重复的标准化 URL 或标题加日期标识，占全部返回记录的比例。 情绪字段覆盖率 暴露情绪值的相关记录占全部相关记录的比例；它衡量可用性，不衡量情绪准确率。 请求语言命中率 匹配请求语言的记录，占语言明确或可检测记录的比例；始终同时展示观察分母。 请求国家命中率 在指定国家的案例中，匹配请求国家的记录占具有可观察国家元数据记录的比例，并始终展示观察分母。 延迟中位数 已完成实时观察的端到端耗时中位数；这是本轮观察值，不是 SLA。 严格 Ticker 市场通过 真实市场只有在两轮中均通过供应商专用 ticker 或 symbol 参数返回目标公司相关新闻时才通过。 负向对照通过 无效查询或 ticker 不得返回文章；无关兜底结果属于失败，不属于成功。 ## The published evaluation standard The scoring unit is one observation: one product, one applicable case, and one live round. A cell is one product-case pair; an API invocation is one physical request. N/A means the inspected contract cannot express the request. Not qualified means applicable tests completed but the best supplier tool did not clear the scenario gate. A scenario ranking requires at least 50% processable responses, relevant results in at least 50% of applicable rounds, and contract coverage for at least 50% of fixed cases. The multi-market ticker ranking separately requires at least 50% processable responses and repeatable relevant results in five or more of nine markets. This is a controlled and reproducible product-selection benchmark, not an academic study or long-term SLA certification. Two rounds show repeatability during the test window, not monthly reliability. ## 公开评价标准 评分单位是一条观察：一个产品、一个适用案例和一轮实时测试。一个单元代表一个产品—案例组合；一次 API invocation 才是一次物理请求。N/A 表示所检查的端点契约无法表达该请求；未达标表示适用测试已完成，但该供应商最好的工具没有越过场景门槛。 进入场景榜单必须同时满足：可处理响应率至少 50%、至少 50% 的适用轮次返回相关结果、端点契约覆盖至少 50% 的固定案例。多市场 Ticker 榜单另要求可处理响应率至少 50%，并在九个市场中的至少五个市场重复返回相关结果。 这是一项受控、可复现的产品选型基准，不是学术研究或长期 SLA 认证。两轮测试只能证明测试窗口内的可重复性。 ![Buyer-scenario qualification across seven financial news API suppliers in the August 2026 live benchmark](https://qveris.ai/guides/financial-news-api-for-ai-agents/images/financial-news-benchmark-scenarios.png) Supplier Global company discovery China/HK company news China specialist news Macro/cross-asset Multi-market ticker Brave Not qualified — Brave News Search, 75% (4) N/A N/A Brave News Search — Qualified, 83% (6) N/A Caidazi Caidazi Hybrid Search — Qualified, 67% (18) Caidazi WeChat Search — Qualified, 100% (4) Caidazi Hybrid Search V2 — Qualified, 100% (6) Caidazi Hybrid Search — Qualified, 50% (12) N/A EODHD EODHD Financial News — Qualified, 78% (18) EODHD Financial News — Qualified, 100% (4) N/A N/A EODHD Financial News — Qualified, 7/9 markets Finlight Not qualified — Finlight Financial News, 50% (4) N/A N/A Finlight Financial News — Qualified, 67% (6) Finlight Financial News — Qualified, 9/9 markets Gildata Gildata Public Opinion — Qualified, 78% (18) Gildata Stock News — Qualified, 100% (4) Gildata Public Opinion — Qualified, 100% (6) Gildata Public Opinion — Qualified, 83% (12) N/A Linkup Linkup Search — Qualified, 100% (18) Linkup Search — Qualified, 100% (4) N/A Linkup Search — Qualified, 100% (12) N/A NewsAPI Not qualified — NewsAPI Everything, 38% (16) Not qualified — NewsAPI Everything, 0% (4) N/A NewsAPI Everything — Qualified, 80% (10) N/A 供应商 全球公司发现 中国/香港公司新闻 中国专业新闻 宏观与跨资产 多市场 Ticker Brave 未达标 — Brave News Search, 75% (4) N/A N/A Brave News Search — 合格, 83% (6) N/A Caidazi Caidazi Hybrid Search — 合格, 67% (18) Caidazi WeChat Search — 合格, 100% (4) Caidazi Hybrid Search V2 — 合格, 100% (6) Caidazi Hybrid Search — 合格, 50% (12) N/A EODHD EODHD Financial News — 合格, 78% (18) EODHD Financial News — 合格, 100% (4) N/A N/A EODHD Financial News — 合格, 7/9 个市场 Finlight 未达标 — Finlight Financial News, 50% (4) N/A N/A Finlight Financial News — 合格, 67% (6) Finlight Financial News — 合格, 9/9 个市场 Gildata Gildata Public Opinion — 合格, 78% (18) Gildata Stock News — 合格, 100% (4) Gildata Public Opinion — 合格, 100% (6) Gildata Public Opinion — 合格, 83% (12) N/A Linkup Linkup Search — 合格, 100% (18) Linkup Search — 合格, 100% (4) N/A Linkup Search — 合格, 100% (12) N/A NewsAPI 未达标 — NewsAPI Everything, 38% (16) 未达标 — NewsAPI Everything, 0% (4) N/A NewsAPI Everything — 合格, 80% (10) N/A ## Best observed fits by agent workflow Use the workflow-specific result, not a universal score. Linkup led global company discovery and macro/cross-asset repeatability. Finlight led structured multi-market ticker news. Gildata, EODHD, Linkup, and Caidazi exposed different strengths in China/HK company or Chinese specialist use cases. Agent workflow Best observed fit Why Structured multi-market ticker news Finlight Financial News 24/24 processable responses; relevant results in both rounds for 9/9 markets; all three negative and dialect controls passed Global company discovery Linkup Search Relevant results in 18/18 observations across all nine fixed company cases China/HK company news Gildata Stock News, EODHD, and Linkup Each returned relevant results in 4/4 applicable China/HK observations Chinese specialist news Caidazi Hybrid Search V2 and Gildata Public Opinion Both returned relevant results in 6/6 observations and reached 100% top-5 relevance Macro and cross-asset discovery Linkup Search Relevant results in 12/12 observations; NewsAPI followed at 8/10 English discovery Brave News Search Relevant results in 5/6 macro and cross-asset observations within its tested scope Multi-market ticker alternative EODHD Financial News Relevant results repeated in 7/9 markets ## 不同 Agent 工作流的实测最佳适配 应按工作流选择，而不是寻找一个通用总分。Linkup 在全球公司发现与宏观/跨资产重复性方面领先；Finlight 在结构化多市场 Ticker 新闻方面领先；Gildata、EODHD、Linkup 与 Caidazi 在中国/香港公司新闻或中文专业新闻场景中各有不同优势。 Agent 工作流 实测最佳适配 原因 结构化多市场 Ticker 新闻 Finlight Financial News 24/24 响应可处理；9/9 市场两轮均返回相关结果；三个负向及 ticker 方言对照全部通过 全球公司发现 Linkup Search 九个固定公司案例的 18/18 次观察均返回相关结果 中国/香港公司新闻 Gildata Stock News, EODHD, and Linkup 三者在 4/4 次适用中国/香港观察中均返回相关结果 中文专业新闻 Caidazi Hybrid Search V2 and Gildata Public Opinion 两者在 6/6 次观察中均返回相关结果，Top-5 相关度均为 100% 宏观与跨资产发现 Linkup Search 12/12 次观察均返回相关结果；NewsAPI 为 8/10 英文新闻发现 Brave News Search 在测试范围内的 5/6 次宏观与跨资产观察中返回相关结果 多市场 Ticker 备选 EODHD Financial News 在 7/9 个市场中重复返回相关结果 ![Global company discovery live expected-response rate and top-five relevance by financial news supplier](https://qveris.ai/guides/financial-news-api-for-ai-agents/images/financial-news-benchmark-global-company.png) ## Scenario-level trade-offs ### China/HK company news EODHD and Gildata Stock News were highly relevant in the fixed company cases. Caidazi exposed richer observable source and core-field data. A zero source count means the normalized response exposed no usable source value; it does not mean the supplier ingested zero publishers. Product Relevant rounds Top-5 relevance Fresh records Core fields Sources/non-empty run Duplicate rate Sentiment on relevant rows Median latency Invalid control Caidazi WeChat Search 4/4 60% 100% 100% 5.0 0% 0% 2,388 ms 0/2 EODHD Financial News 4/4 100% 0% 60% 0.0 0% 100% 2,585 ms 0/2 Gildata Stock News 4/4 95% 25% 60% 0.0 0% 0% 17,621 ms 2/2 Linkup Search 4/4 60% 0% 40% 0.0 0% 0% 4,177 ms 0/2 ### Chinese specialist news Caidazi Hybrid Search V2 and Gildata Public Opinion both returned relevant results in all six observations and reached 100% top-five relevance. Caidazi was faster and more structurally complete in this run. Product Observations Relevant rounds Top-5 relevance Fresh records Core fields Sources/non-empty run Duplicate rate Median latency Caidazi Hybrid Search V2 6 6/6 100% 27% 100% 4.0 0% 2,547 ms Gildata Public Opinion 6 6/6 100% 42% 43% 0.0 0% 25,013 ms ### Macro and cross-asset news Linkup had the strongest relevant-result repeatability. An invalid-control failure means the endpoint returned fallback content for the deliberately nonexistent query; agents should add their own relevance guard. Product Observations Relevant rounds Top-5 relevance Fresh records Core fields Sources/non-empty run Duplicate rate Median latency Invalid control Brave News Search 6 5/6 96% 0% 60% 0.0 0% 2,443 ms 0/2 Caidazi Hybrid Search 12 6/12 43% 60% 81% 4.5 7% 3,938 ms 0/2 Finlight Financial News 6 4/6 90% 100% 100% 2.0 0% 2,091 ms 2/2 Gildata Public Opinion 12 10/12 90% 0% 60% 0.0 0% 14,603 ms 2/2 Linkup Search 12 12/12 100% 0% 40% 0.0 0% 4,560 ms 0/2 NewsAPI Everything 10 8/10 70% 39% 100% 2.6 0% 1,922 ms 2/2 ## 分场景权衡 ### 中国/香港公司新闻 EODHD 与 Gildata Stock News 在固定公司案例中相关性很高；Caidazi 暴露了更丰富的可观察来源与核心字段。来源数为零只表示标准化响应没有可用来源字段，不代表供应商没有抓取发布者。 产品 相关轮次 Top-5 相关度 新鲜记录 核心字段 每个非空轮次的来源数 重复率 相关结果的情绪覆盖 延迟中位数 无效输入对照 Caidazi WeChat Search 4/4 60% 100% 100% 5.0 0% 0% 2,388 ms 0/2 EODHD Financial News 4/4 100% 0% 60% 0.0 0% 100% 2,585 ms 0/2 Gildata Stock News 4/4 95% 25% 60% 0.0 0% 0% 17,621 ms 2/2 Linkup Search 4/4 60% 0% 40% 0.0 0% 0% 4,177 ms 0/2 ### 中文专业新闻 Caidazi Hybrid Search V2 与 Gildata Public Opinion 在六次观察中全部返回相关结果，Top-5 相关度均为 100%。本轮测试中 Caidazi 更快，结构也更完整。 产品 观察数 相关轮次 Top-5 相关度 新鲜记录 核心字段 每个非空轮次的来源数 重复率 延迟中位数 Caidazi Hybrid Search V2 6 6/6 100% 27% 100% 4.0 0% 2,547 ms Gildata Public Opinion 6 6/6 100% 42% 43% 0.0 0% 25,013 ms ### 宏观与跨资产新闻 Linkup 的相关结果重复性最强。无效输入对照失败意味着端点对故意构造的不存在查询返回了兜底内容，Agent 必须增加自己的相关性校验。 产品 观察数 相关轮次 Top-5 相关度 新鲜记录 核心字段 每个非空轮次的来源数 重复率 延迟中位数 无效输入对照 Brave News Search 6 5/6 96% 0% 60% 0.0 0% 2,443 ms 0/2 Caidazi Hybrid Search 12 6/12 43% 60% 81% 4.5 7% 3,938 ms 0/2 Finlight Financial News 6 4/6 90% 100% 100% 2.0 0% 2,091 ms 2/2 Gildata Public Opinion 12 10/12 90% 0% 60% 0.0 0% 14,603 ms 2/2 Linkup Search 12 12/12 100% 0% 40% 0.0 0% 4,560 ms 0/2 NewsAPI Everything 10 8/10 70% 39% 100% 2.6 0% 1,922 ms 2/2 ## Strict ticker lookup by market Ticker support is not keyword recall. A product may find an article containing “Toyota” yet fail a strict `7203.T` lookup. Normalize each user symbol to the provider's native dialect before calling the endpoint. Finlight completed 24/24 strict-ticker calls and passed US, HK, CN, JP, DE, FR, BR, IN, and ES in both rounds. The negative controls—an invalid ticker, `600519.SS`, and secondary listing `PETR4`—returned empty in both rounds. EODHD passed seven of nine markets. Product Markets passing both rounds Edge-case behavior Finlight Financial News US, HK, CN, JP, DE, FR, BR, IN, ES Invalid ticker, 600519.SS, and secondary listing PETR4 returned empty in both rounds; primary tickers passed EODHD Financial News US, HK, CN, DE, FR, BR, ES JP, IN, and invalid ticker returned provider errors ## 按市场进行严格 Ticker 查询 Ticker 支持不等于关键词召回。产品可能找到包含 “Toyota” 的文章，却无法完成严格的 `7203.T` 查询。调用前必须把用户输入的代码转换成供应商的原生 Ticker 方言。 Finlight 完成 24/24 次严格 Ticker 调用，并在 US、HK、CN、JP、DE、FR、BR、IN、ES 九个市场两轮均通过。无效 ticker、`600519.SS` 与次级上市代码 `PETR4` 三个负向对照两轮均为空。EODHD 通过七个市场。 产品 两轮均通过的市场 边界情况表现 Finlight Financial News US, HK, CN, JP, DE, FR, BR, IN, ES 无效 ticker、600519.SS 和次级上市代码 PETR4 两轮均为空；主 ticker 通过 EODHD Financial News US, HK, CN, DE, FR, BR, ES JP、IN 和无效 ticker 返回供应商错误 ![Two-round ticker-filtered relevant financial news results across nine listing markets](https://qveris.ai/guides/financial-news-api-for-ai-agents/images/financial-news-benchmark-ticker-markets.png) ## Which languages do the news APIs support? Language coverage has three layers: what the vendor documents, what the current integration lets an agent request, and what the fixed live cases returned. Do not infer an undocumented inventory from one successful result. Linkup returned relevant requested-language results in de, en, es, fr, ja, pt, and zh, but exposes no explicit language filter in the tested contract. NewsAPI exposed 14 language codes, while only English repeated in the fixed company cases. Finlight's native API documents one ISO 639-1 filter per request; the current QVeris connector does not expose it. A separate native correction suite repeated Chinese results for zh, zh-CN, and zh-Hans. Product Documented or exposed language control Repeatably observed requested languages Finlight Financial News Native API accepts one ISO 639-1 language per request and normalizes aliases; current QVeris integration does not expose the parameter Connector: en; native correction suite: zh, including zh-CN and zh-Hans aliases Linkup Search No explicit language filter in the tested contract de, en, es, fr, ja, pt, zh NewsAPI Everything 14 exposed codes: ar, de, en, es, fr, he, it, nl, no, pt, ru, sv, ud, zh en Brave News Search Current integration enumerates ar, bg, bn, ca, en, eu en ## 这些新闻 API 支持哪些语言？ 语言覆盖必须分三层描述：供应商文档声明了什么、当前集成允许 Agent 请求什么、固定实时案例实际返回了什么。不能依据一次成功结果推断未公开的语言库存。 Linkup 在 de、en、es、fr、ja、pt、zh 七种请求语言中均重复返回相关结果，但测试契约没有显式语言筛选参数。NewsAPI 暴露 14 个语言代码，但固定公司案例中只有英文重复命中。Finlight 原生 API 文档支持每次请求一个 ISO 639-1 语言；当前 QVeris 连接器尚未暴露该参数。独立原生纠偏测试中，zh、zh-CN 与 zh-Hans 均重复返回中文结果。 产品 文档或集成暴露的语言控制 重复观察到的请求语言 Finlight Financial News 原生 API 每次请求接受一种 ISO 639-1 语言并规范化别名；当前 QVeris 集成未暴露该参数 连接器：en；原生纠偏测试：zh，包括 zh-CN 与 zh-Hans 别名 Linkup Search 测试契约中没有显式语言筛选参数 de, en, es, fr, ja, pt, zh NewsAPI Everything 暴露 14 个代码：ar、de、en、es、fr、he、it、nl、no、pt、ru、sv、ud、zh en Brave News Search 当前集成枚举 ar、bg、bn、ca、en、eu en ![Requested-language coverage observed in QVeris financial news connector tests](https://qveris.ai/guides/financial-news-api-for-ai-agents/images/financial-news-benchmark-languages.png) ## Market-targeted and requested-language results Each applicable tool-case cell ran twice. A percentage is passing rounds divided by two. A round passes only when the response has the expected shape, returns at least one relevant result, and all language-observable top-five rows match the requested language. It is not an article-level language-accuracy score or a permanent coverage claim. Country metadata was observable in all returned Finlight rows from its two applicable default-English company cases and matched the requested country. It was not observable in the selected rows for the other six suppliers. Missing metadata is N/A, not evidence of zero coverage. Supplier Country metadata observable in returned rows Requested-country match when observable Brave 0% N/A Caidazi 0% N/A EODHD 0% N/A Finlight 100% 100% Gildata 0% N/A Linkup 0% N/A NewsAPI 0% N/A ## 面向市场和请求语言的结果 每个适用的工具—案例单元都运行两轮。百分比等于通过轮次数除以二；一轮只有在响应形状符合预期、至少返回一条相关结果、且所有可检测语言的 Top-5 记录都匹配请求语言时才通过。这不是文章级语言准确率，也不是永久覆盖声明。 在 Finlight 两个适用的默认英文公司案例中，全部返回记录都能观察到国家元数据并匹配请求国家；其余六家选定结果中无法观察到该字段。缺失元数据应记为 N/A，不能解释成零覆盖。 供应商 返回结果中可观察的国家元数据 元数据可观察时的请求国家匹配率 Brave 0% N/A Caidazi 0% N/A EODHD 0% N/A Finlight 100% 100% Gildata 0% N/A Linkup 0% N/A NewsAPI 0% N/A ![Market-targeted company relevance and requested-language pass rate across seven suppliers](https://qveris.ai/guides/financial-news-api-for-ai-agents/images/financial-news-benchmark-market-language.png) ## Selection guidance - **Choose Finlight** when the agent starts from a ticker and needs structured entities, sentiment labels, categories, provenance, and clean negative-control behavior. - **Choose EODHD** when broad exchange-coded ticker coverage matters and the agent can handle provider errors explicitly. - **Choose Linkup** when multilingual recall and citations matter more than a strict news-feed contract. - **Choose Brave** for English discovery within its tested scope. - **Choose NewsAPI** for a familiar keyword API and explicit language selection, while recognizing that declared language options do not guarantee relevant local-market results. - For China/HK company workflows, compare Gildata Stock News, EODHD, Linkup, and Caidazi WeChat Search against the response shape you actually need. - For Chinese fund, industry, and organization workflows, compare Caidazi Hybrid Search V2 with Gildata Public Opinion. ## 选择建议 - 当 Agent 从 Ticker 出发，并需要结构化实体、情绪标签、分类、来源证据和干净的负向对照表现时，选择 **Finlight**。 - 当广泛的交易所编码 Ticker 覆盖更重要，且 Agent 能明确处理供应商错误时，选择 **EODHD**。 - 当多语言召回与引用比严格新闻 Feed 契约更重要时，选择 **Linkup**。 - 在本次测试范围内需要英文发现时，可考虑 **Brave**。 - 需要熟悉的关键词 API 与显式语言选项时，可考虑 **NewsAPI**，但声明的语言选项不保证本地市场结果相关。 - 中国/香港公司工作流应按目标响应结构比较 Gildata Stock News、EODHD、Linkup 与 Caidazi WeChat Search。 - 中文基金、行业和组织工作流应比较 Caidazi Hybrid Search V2 与 Gildata Public Opinion。 ## Submit a provider or request a retest提交供应商或申请复测 Benchmark inclusion and ranking cannot be purchased. Providers may submit a comparable discovery or ticker-news endpoint, document market and language controls, and provide a test credential or QVeris integration.基准收录与排名不能付费购买。供应商可以提交可比较的新闻发现或 Ticker 新闻端点、记录市场与语言控制，并提供测试凭证或 QVeris 集成。 [Check the Provider Hub查看 Provider Hub](https://qveris.ai/providers) [Request a benchmark retest申请基准复测](mailto:support@qveris.ai?subject=Financial%20News%20Benchmark) ## Limitations and retest checklist This is a dated live observation, not a permanent claim about a provider's full inventory. Results vary with the news cycle, plan, licensing, indexing delay, and integration schema. Empty results show what the fixed query returned in the test window; they do not prove contractual non-support. Rerun quarterly and after any material endpoint, source, plan, schema, language, or ticker-routing change. Keep cases stable so score changes reflect the product rather than a rewritten benchmark. - Exact endpoint and stable schema - Supported ticker formats and exchanges - Language codes and article-level versus source-level filtering - Country-field semantics - Source, licensing, and freshness constraints - Invalid-input and empty-result behavior ## 局限与复测清单 这是一份带日期的实时观察，不是对供应商全部库存的永久声明。结果会随新闻周期、套餐、授权、索引延迟与集成 Schema 变化。空结果只说明固定查询在测试窗口内没有返回相关文章，不能证明契约层面不支持该市场。 建议每季度复测，并在端点、来源、套餐、Schema、语言或 Ticker 路由发生实质变化后复测。保持案例稳定，才能让分数变化反映产品变化，而不是测试题变化。 - 精确端点与稳定 Schema - 支持的 Ticker 格式与交易所 - 语言代码，以及筛选作用于文章还是来源 - 国家字段语义 - 来源、授权与新鲜度约束 - 无效输入与空结果行为 ## FAQ ### Can I search financial news by stock ticker?可以按股票 Ticker 搜索金融新闻吗？ Yes. Several tested integrations exposed dedicated ticker or symbol inputs, but only Finlight and EODHD met the published multi-market ranking gate. Normalize the user's symbol to each provider's dialect.可以。多个测试集成都暴露了专用 ticker 或 symbol 参数，但只有 Finlight 与 EODHD 达到公开的多市场排名门槛。应把用户代码规范化为各供应商的方言。 ### Which API had the broadest observed language coverage?哪一个 API 的实测语言覆盖最广？ Linkup repeated relevant results in all seven requested languages, but it is a Web retrieval product rather than a strict structured financial news feed. Among explicit news APIs in the tested integrations, English was the only language repeated in the fixed company cases.Linkup 在七种请求语言中都重复返回相关结果，但它属于 Web 检索产品，并非严格的结构化金融新闻 Feed。在测试集成的明确新闻 API 中，固定公司案例只有英文重复命中。 ### How often should the benchmark be rerun?这个基准应该多久重跑一次？ Quarterly, and after any material endpoint, source, plan, schema, language, or ticker-routing change. Keep the cases stable.每季度一次，并在端点、来源、套餐、Schema、语言或 Ticker 路由发生重大变化后重跑；测试案例应保持稳定。 ## Official and related sources官方资料与相关指南 ### Supplier websites and documentation供应商官网与文档 ### Related QVeris guidesQVeris 相关指南