Real-Time Financial News API
Sources and Latency Compared实时金融新闻 API
来源、延迟与授权对比
A fast headline feed is not automatically complete, licensed for display, or suitable for automated decisions. Compare the whole event pipeline.
快速标题流不等于覆盖完整、允许展示或适合自动决策。应比较整个事件管线。
TL;DR
Finnhub's free plan documents one year of company news plus new updates, while its real-time news WebSocket is premium.
Alpha Vantage combines news search with ticker/topic filters and sentiment, subject to free-key rate limits.
Webhook delivery reduces polling waste, but production systems still need signatures, retries, idempotency, and replay.
Compare event latency, source provenance, full-text rights, entity precision, archive depth, and delivery guarantees separately.
Finnhub 免费套餐说明包含一年公司新闻和新增更新，但实时新闻 WebSocket 属于 Premium。
Alpha Vantage 将新闻搜索、代码/主题筛选和情绪结合，但受免费 Key 请求限制。
Webhook 可减少轮询浪费，但生产系统仍需签名、重试、幂等和重放。
分别比较事件延迟、来源、全文权利、实体准确率、历史深度和交付保证。
What “free” and “real-time” mean for this API这个 API 中“免费”与“实时”分别意味着什么
Real-time financial news is a pipeline, not a single timestamp. A company publishes a filing or press release, a source ingests it, the provider normalizes and tags it, the API delivers it, and your application processes it. Measuring only API response time ignores most of the delay.
Coverage also has layers: general market headlines, company news, regulatory filings, investor-relations releases, analyst actions, macro events, and licensed publisher content. A feed can be fast for press releases yet weak for editorial news, or broad in headlines while offering only links rather than licensed full text.
AI systems need more than sentiment. They need stable article IDs, canonical URLs, source and publication time, related tickers, duplicate clusters, corrected-story handling, and an evidence path back to the original item.
实时金融新闻是一条管线，不是单个时间戳。公司先发布公告或新闻稿，来源系统采集，供应商再标准化和打标签，API 交付后应用处理。只测接口响应时间，会忽略大部分延迟。
覆盖也有多个层次：市场综合新闻、公司新闻、监管公告、投资者关系稿、评级变化、宏观事件和有版权的媒体内容。某个源可能新闻稿很快，却缺少编辑新闻；也可能标题很广，但只给链接而没有全文权利。
AI 系统不能只要情绪分数，还需要稳定文章 ID、规范 URL、来源与发布时间、关联代码、去重簇、修订处理，以及回到原始内容的证据链。
Provider comparison: access, fields, and limits供应商对比：入口、字段与限制
| Provider供应商 | Free entry免费入口 | Delivery交付方式 | Important boundary重要边界 |
|---|---|---|---|
| FinnhubFinnhub | Company news and market news endpoints公司新闻与市场新闻端点 | REST updates; premium WebSocket for real-time newsREST 更新；实时新闻 WebSocket 为 Premium | North American scope and plan-specific history北美范围与套餐历史深度 |
| Alpha VantageAlpha Vantage | Free API key免费 API Key | REST news and sentimentREST 新闻与情绪 | Rate limits and licensed content fields vary请求限制与内容字段有差异 |
| FinlightFinlight | Developer evaluation by current plan按当前套餐提供开发评估 | REST plus documented webhooksREST 与文档化 Webhook | Confirm plan, source, and full-text rights核验套餐、来源和全文权利 |
| Twelve Data press releasesTwelve Data 新闻稿 | Basic plan and aboveBasic 及以上 | REST filtered by symbol, exchange, date, language按代码、交易所、日期和语言筛选的 REST | Press releases are not the same as editorial news新闻稿不等于编辑新闻 |
Provider-by-provider notes: where the free tier stops逐供应商深挖：免费层在哪里结束
Finnhub's plan table and API documentation separate two useful concepts. Company-news REST access can include recent history and new updates on the free plan, while the low-latency news WebSocket is labeled premium. A polling prototype can therefore be free without being equivalent to a continuous premium stream.
Alpha Vantage is useful when a prototype needs ticker and topic filters plus provider-generated sentiment in one response. Validate article coverage, duplicate behavior, timestamps, and free-key throughput. Treat sentiment as enrichment and retain the source URL and original text evidence for any conclusion.
Finlight documents webhooks for event-driven delivery, which can be a better production pattern than aggressive polling. The engineering burden moves to signature verification, idempotent consumers, retries, dead-letter storage, and replay. Confirm which sources and fields are included in the current plan before promising real-time coverage.
Twelve Data's press-release endpoint is valuable for official corporate announcements filtered by symbol, exchange, date, or language. It should be classified as first-party event content rather than a complete editorial-news replacement. A robust research product often combines releases, filings, and licensed journalism.
Finnhub 的套餐表和文档区分了两个概念：免费层可通过公司新闻 REST 获取近期历史和新增更新，而低延迟新闻 WebSocket 标为 Premium。因此免费轮询原型并不等于连续的高级实时流。
当原型需要代码/主题筛选和供应商情绪字段时，Alpha Vantage 很方便。但要验证文章覆盖、重复、时间戳和免费 Key 吞吐量；情绪只能作为增强信息，结论仍需保留来源 URL 和原文证据。
Finlight 提供事件驱动 Webhook 文档，生产中可能比高频轮询更合适。但工程责任会转向签名验证、幂等消费、重试、死信存储和重放；承诺实时覆盖前仍要确认当前套餐包含哪些来源和字段。
Twelve Data 新闻稿端点适合按代码、交易所、日期或语言获取官方公司公告，但它属于一手事件内容，不能等同于完整编辑新闻。稳健研究产品通常会组合新闻稿、监管公告和有版权的新闻媒体。
The fields that decide whether the data is usable决定数据是否可用的关键字段
Track source publication time, provider ingestion time, API availability time, and application processing time.
Retain publisher, canonical URL, document type, language, and whether content is original, syndicated, or a press release.
Ticker tagging must handle subsidiaries, competitors, ETFs, ambiguous names, and articles mentioning a company only in passing.
Polling, WebSocket, and webhooks have different failure modes. Define retry, ordering, replay, and duplicate behavior.
记录来源发布时间、供应商采集时间、API 可用时间和应用处理时间。
保留发布者、规范 URL、文档类型、语言，以及内容是原创、转载还是新闻稿。
代码标注要处理子公司、竞争对手、ETF、歧义名称和仅顺带提及公司的文章。
轮询、WebSocket 和 Webhook 的故障模式不同，应定义重试、顺序、重放和重复处理。
Which option fits your workload?哪一种选择适合你的工作负载？
Free REST updates can work when a minute-scale poll is acceptable. Deduplicate stories and store the last successful cursor or publication time.
Prefer WebSocket or webhook delivery with measured source-to-application latency, reconnect or retry behavior, and a replay endpoint.
Keep article-level evidence and model version, cluster syndications, and show distribution rather than collapsing a day into one unexplained score.
Require source class, freshness, entity confidence, corroboration, policy approval, and risk limits. One headline or one sentiment score should never authorize an irreversible action.
如果可以接受分钟级轮询，免费 REST 更新就能工作；需要去重并保存最后成功游标或发布时间。
优先使用 WebSocket 或 Webhook，测量来源到应用总延迟，并设计重连/重试和重放端点。
保留文章级证据与模型版本，对转载聚类，展示情绪分布，而不是把一天压成一个无法解释的分数。
要求来源类别、新鲜度、实体置信度、交叉验证、策略批准和风险限制；单条标题或单个情绪分数不能授权不可逆动作。
A practical evaluation and implementation playbook可执行的评估与落地方法
Use earnings, M&A rumors, regulatory filings, analyst actions, and corrected stories to measure recall, precision, and latency.
Syndicated stories and minor headline edits should cluster without hiding genuinely new facts.
Verify signatures, store delivery IDs, make consumers idempotent, retry with backoff, and provide a replay path.
Require source class, confidence, freshness, and corroboration rules before news triggers a trade or external action.
用财报、并购传闻、监管公告、评级变化和修订新闻测试召回、准确和延迟。
转载和轻微标题修改应聚类，但不能隐藏真正的新事实。
验证签名、保存交付 ID、消费者保持幂等、退避重试，并提供重放路径。
新闻触发交易或外部动作前，应校验来源类别、置信度、新鲜度和交叉验证规则。
Common mistakes that make a free API expensive让免费 API 变昂贵的常见错误
A one-minute poll has a built-in delay even if the upstream item is available instantly.
A score should link to the article and model/version that produced it.
Receiving a headline and URL does not grant permission to store or republish the article body.
Ten syndications of one press release are one event, not ten independent confirmations.
即使上游立即可用，一分钟轮询也自带延迟。
情绪分数应关联原文以及产生分数的模型和版本。
收到标题和 URL 不代表可以存储或再发布正文。
同一新闻稿的十次转载仍是一个事件，不是十次独立确认。
Where QVeris fits in the workflowQVeris 在工作流中的位置
QVeris is a capability discovery and routing layer for agents. It can help an agent find, inspect, and call financial capabilities, but it does not replace the underlying data provider's license, entitlement, or source-of-truth obligations.
QVeris 是面向 Agent 的能力发现与路由层，可帮助 Agent 找到、检查并调用金融能力，但不会替代底层数据供应商的许可、权限或真值来源责任。
- QVeris can expose news, filings, and market-data capabilities through a common discovery and inspection workflow for agents.
- Each answer should preserve article ID, source, canonical URL, publication time, retrieval time, related entities, and provider.
- Content licenses and automated-action policy remain separate controls; a capability route does not remove either obligation.
- QVeris 可把新闻、公告和行情能力放进统一的 Agent 发现与检查流程。
- 每个回答应保留文章 ID、来源、规范 URL、发布时间、获取时间、相关实体和供应商。
- 内容许可和自动动作策略仍是独立控制；能力路由不会消除这些义务。
FAQ
There is no universal winner. Measure source-to-application latency for the event types and companies you care about.
Free REST updates exist, but low-latency WebSockets, premium publishers, full text, and commercial rights often require paid plans.
WebSocket fits continuous subscriptions; webhooks fit server-to-server events. Both need reconnect or retry and replay design.
Not from one score alone. Require provenance, freshness, entity checks, corroboration, risk limits, and human or policy approval.
没有统一答案，应针对目标事件类型和公司测量从来源到应用的总延迟。
存在免费 REST 更新，但低延迟 WebSocket、优质媒体、全文和商业权利通常需要付费。
WebSocket 适合持续订阅，Webhook 适合服务器事件；两者都需要重连或重试与重放设计。
不能只依赖一个分数，应要求来源、新鲜度、实体检查、交叉验证、风险限制及人工或策略批准。
Sources, methodology, and update policy来源、方法与更新政策
This guide compares public documentation and evaluates providers by workload, data freshness, structure, delivery, legal use, and upgrade path. It does not treat a marketing label as proof of real-time coverage.
Provider features and commercial terms can change. Official documentation linked below is the source of truth; run a representative data test and obtain appropriate rights before production deployment.
本文依据公开文档，按工作负载、数据新鲜度、结构、交付、合法用途和升级路径比较供应商，不把营销标签当作实时覆盖的证明。
供应商功能和商业条款可能变化。以下官方文档是最终依据；生产部署前应运行代表性数据测试并取得适当权利。