Meta AI定价拆解：用数据换折扣的“广告模式”与隐私溢价
The Ads Model for Prompts Vertically Integrates AI
提供了具体的API定价参数与隐私溢价计算逻辑，帮助开发者评估数据资产价值及合规成本，是理解当前AI底层经济模型的实操参考。
Meta launched two things yesterday : a state-of-the-art model & a new pricing system for foundation models.1 Muse Spark propels US open source models to the frontier. Meanwhile, the pricing system resets the industry’s economics.
Meta 昨天发布了两项内容：一款最先进的模型以及针对基础模型的新定价体系。1 Muse Spark 将美国开源模型推向了前沿。与此同时，该定价体系重塑了行业的经济模式。
For thirty years, enterprise software operated on strict licensing fees with a guarantee of total privacy & zero data retention.2 Consumer technology operated on the opposite principle : free software in exchange for behavioral data. Consumers trade queries for convenience, but enterprises fiercely protect their intellectual property.
三十年来，企业软件一直实行严格的许可费制度，并承诺完全隐私且零数据留存。2 消费级技术则遵循相反的原则：以行为数据换取免费软件。消费者用查询权换取便利性，但企业会 fiercely（激烈地）保护其知识产权。
Now that model is coming to AI with a twist.
如今，这种模式正带着变数进入 AI 领域。
Meta’s new pricing makes the barter explicit through a two-tier pricing structure with & without privacy :3
Meta 的新定价通过包含与不包含隐私的两层定价结构，使这种易货交易变得明确：3
- The Standard Tier (muse-spark-1.3) : $1.25/m input tokens & $4.25/m output tokens. Your prompts & completions are never used to train Meta’s foundation models.
- The Contributor Tier (muse-spark-1.3-contributor) : $0.10/m input tokens & $0.20/m output tokens. In exchange for a 92% discount on input & a 95% discount on output, Meta retains the right to train future models on your data.
- 标准层（muse-spark-1.3）：输入 token 每百万 $1.25，输出 token 每百万 $4.25。你的提示词和补全内容永远不会被用于训练 Meta 的基础模型。
- 贡献者层（muse-spark-1.3-contributor）：输入 token 每百万 $0.10，输出 token 每百万 $0.20。作为对输入折扣 92% 和输出折扣 95% 的交换，Meta 保留使用你的数据训练未来模型的权利。
No other foundation model provider currently offers this kind of explicit barter on its API. This is the ads model coming to AI.
目前没有其他基础模型提供商在其 API 上提供这种明确的易货交易。这是广告模式在 AI 领域的延伸。
As Michael Mauboussin writes in Expectations Investing, there is information in prices.4 When Meta charges two radically different prices for the same AI, the difference in price tells us the value of the data.
正如 Michael Mauboussin 在《Expectations Investing》中所写，价格中包含着信息。4 当 Meta 为同一种 AI 收取两种截然不同的价格时，价格差异告诉了我们数据的价值。
At an agentic 30:1 input-to-output ratio,5 the blended standard tier costs $1.35/m tokens, while the contributor tier costs $0.103/m. The difference is a $1.24/m token spread, an effective 92.3% subsidy.
在代理场景下 30:1 的输入输出比下，5 混合后的标准层成本为每百万 token $1.35，而贡献者层成本为每百万 token $0.103。差额为每百万 token $1.24 的价差，相当于 92.3% 的有效补贴。
| Daily Token Volume | Monthly Tokens | Annual Cost (Private / ZDR) | Annual Cost (Data Sharing) | Annual Data Dividend |
|---|---|---|---|---|
| 10m tokens / day | 300m | $4,916 | $377 | $4,539 |
| 100m tokens / day | 3b | $49,157 | $3,768 | $45,389 |
| 1b tokens / day | 30b | $491,573 | $37,677 | $453,895 |
| 每日 Token 量 | 每月 Token 量 | 年度成本（私有 / ZDR） | 年度成本（数据共享） | 年度数据分红 |
|---|---|---|---|---|
| 1000 万 token/天 | 3 亿 | $4,916 | $377 | $4,539 |
| 1 亿 token/天 | 30 亿 | $49,157 | $3,768 | $45,389 |
| 10 亿 token/天 | 300 亿 | $491,573 | $37,677 | $453,895 |
For an enterprise processing 1b tokens per day, opting into Zero Data Retention (ZDR) is a $454,000 annual privacy surcharge. That difference is how Meta values incoming customer data : $1.24/m tokens.
对于每天处理 10 亿 token 的企业而言，选择零数据留存（ZDR）是一项每年 45.4 万美元的隐私附加费。这一差额正是 Meta 对 incoming customer data（流入的客户数据）的估值：每百万 token $1.24。
This unbundles the $20 monthly consumer subscription. Labs absorbed compute losses on flat-rate consumer plans because default terms granted training rights, an implicit data subsidy that Meta has now formalized per token.6
这将每月 20 美元的消费者订阅进行了拆分。实验室在固定价格的消费者计划中吸收了计算亏损，因为默认条款授予了训练权，这是一种隐性的数据补贴，而 Meta 现在已将其按 token 形式正式化。6
Why surrender 92% of inference revenue? It is not out of altruism.
为什么要放弃 92% 的推理收入？这并非出于利他主义。
The public web has been exhaustively crawled ; frontier gains now come from post-training, reinforcement learning from AI feedback, & user usage patterns. Estimates place the market for training data & human labeling at $10b in annual revenue.7
公共网络已被彻底爬取；前沿突破现在来自训练后阶段、基于 AI 反馈的强化学习以及用户使用模式。估计训练数据与人工标注市场的年收入为 100 亿美元。7
Meta’s pricing model bypasses this intermediary. Just as search & social platforms vertically integrated the digital advertising supply chain by capturing behavioral data directly from users, Meta is vertically integrating the AI data supply chain. Instead of paying labeling vendors to simulate human behavior, Meta turns its inference network into a self-funding data flywheel.
Meta 的定价模型绕过了这一中间环节。正如搜索引擎和社交平台通过直接从用户那里获取行为数据，将数字广告供应链垂直整合一样，Meta 正在将 AI 数据供应链垂直整合。Meta 不再向标注供应商付费以模拟人类行为，而是将其推理网络转化为一个自我资助的数据飞轮。
At $1.24/m tokens of subsidy,8 Meta acquires organic reasoning traces at pennies on the dollar compared to specialized data labs, while undercutting closed foundation models on inference price.
在每百万 token 补贴 1.24 美元的情况下，8 相比专业数据实验室，Meta 以极低的价格获取了有机的推理轨迹，同时在推理价格上低于封闭的基础模型。
Compute is no longer sold simply as an infrastructure utility. It has become a currency traded directly for the training tokens needed to build the next frontier model.
算力不再仅仅作为基础设施服务出售。它已成为一种货币，直接用于交易构建下一代前沿模型所需的训练 token。
Ultimately, this solves the business model for American open source. Just like ads, the barter is simple : subsidized access in exchange for data.
最终，这解决了美国开源领域的商业模式。就像广告一样，这种易货交易很简单：以数据换取补贴访问权限。
- Open Models Tack Toward the Frontier. ↩︎
- Enterprise master services agreements & compliance frameworks (SOC 2, ISO 27001, HIPAA) require zero data retention & prohibit vendor model training on customer data. ↩︎
- Meta Model API, Pricing & Data Terms. ↩︎
- Michael Mauboussin, Expectations Investing. ↩︎
- The Hungry, Hungry AI Model. ↩︎
- At an individual volume of 10m tokens/month, the $1.24/m spread yields a $12/month subsidy. For power users consuming 20m–50m tokens/month, the data subsidy reaches $25–$62/month, matching the consumer subscription discount. ↩︎
- @deedydas on X, Every Single Startup Selling AI Training Data. ↩︎
- At $1.24 per million tokens, Meta’s effective subsidy is $0.0037 per 3,000-token interaction trace. By comparison, specialized human labeling vendors charge $25 to $100+ per hour for domain experts, yielding $5 to $50 per verified reasoning trajectory. ↩︎
- 开源模型转向前沿。↩︎
- 企业主服务协议与合规框架（SOC 2、ISO 27001、HIPAA）要求零数据留存，并禁止供应商对客户数据进行模型训练。↩︎
- Meta 模型 API、定价与数据条款。↩︎
- Michael Mauboussin，《预期投资》。↩︎
- 饥饿的 AI 模型。↩︎
- 在每月 1000 万 token 的个人用量下，每百万 token 1.24 美元的差价产生每月 12 美元的补贴。对于每月消耗 2000 万至 5000 万 token 的重度用户，数据补贴达到每月 25 至 62 美元，这与消费者订阅折扣相当。↩︎
- @deedydas 在 X 平台上的文章《所有销售 AI 训练数据的初创公司》。↩︎
- 按每百万 token 1.24 美元计算，Meta 的有效补贴为每次 3000 token 的交互轨迹 0.0037 美元。相比之下，专业的人工标注供应商对领域专家每小时收费 25 至 100 多美元，每条经过验证的推理轨迹产生 5 至 50 美元的收益。↩︎
更进一步：量化金融体系
看懂新闻只是起点——沿量化金融路径，把它变成能交付的工程能力