OpenAI, Anthropic and xAI Talent Structure Benchmark

01 · 核心结论01 · Key findings

看起来都是工程团队，第二人才重心却完全不同All three look engineering-led, but their secondary talent concentrations are completely different

Engineering 占比只相差 4.4 个百分点；真正的分野是 OpenAI 的产品与模型组合、Anthropic 的 GTM / Customer，以及 xAI 的 Human Data & Evaluation。Engineering shares differ by only 4.4 percentage points; the real split is OpenAI’s product-and-model mix, Anthropic’s GTM / Customer weight, and xAI’s Human Data & Evaluation operation.

9,421

公开职业档案样本Public professional profiles

OpenAI 5,313 · Anthropic 2,612 · xAI 1,496

≈40%

三家公司 Engineering 占比Engineering share at all three companies

40.3%–44.7%

5.83×

xAI Human Data & Evaluation over-index

28.4% 岗位占比28.4% role share

1.58×

Anthropic GTM / Customer over-index

22.6% 岗位占比22.6% role share

01Engineering 占比几乎不是差异Engineering share is barely the differentiator

Engineering 占比介于 40.3% 与 44.7%；相比之下，OpenAI 的 Product / Design 与 GTM / Customer 均为 12.7%，Anthropic 的 GTM / Customer 为 22.6%，xAI 的 Human Data & Evaluation 为 28.4%，区分三家公司应看非 Engineering 配置。Engineering accounts for 40.3% to 44.7%; by contrast, Product / Design and GTM / Customer are both 12.7% at OpenAI, GTM / Customer is 22.6% at Anthropic, and Human Data & Evaluation is 28.4% at xAI, so the non-Engineering mix is the real differentiator.

02xAI 的第二人才重心是规模化评估与数据运营xAI’s distinctive secondary talent concentration is evaluation and data operations

Human Data & Evaluation 占 28.4%、over-index 达 5.83×，而 Research / Models 仅占 2.1%；xAI 的反常之处不是研究岗位更重，而是评估与数据运营规模更大。Human Data & Evaluation accounts for 28.4% with a 5.83× over-index, while Research / Models is only 2.1%; xAI stands out for the scale of its evaluation and data operation, not for a heavier research mix.

03Anthropic 的第二底盘是 GTM，不是 ResearchAnthropic’s second foundation is GTM, not Research

GTM / Customer 占 22.6%，高于 OpenAI 的 12.7% 和 xAI 的 5.6%；同时 Enterprise SaaS 贡献 35.3% 的上一站来源。GTM / Customer reaches 22.6%, versus 12.7% at OpenAI and 5.6% at xAI, while Enterprise SaaS contributes 35.3% of prior-employer sources.

04OpenAI 主要从成熟平台吸收人才，而非以实验室间流动为主OpenAI draws mainly from mature platforms rather than lab-to-lab movement

Enterprise SaaS 与 Big Tech / Platform 合计占上一站来源的 51.7%，AI Lab / Model Co. 与 Academia / Research 合计为 8.4%；结合 Product / Design 1.34× 和 Research / Models 1.26× 的 over-index，成熟技术组织中的产品与模型建设者是更匹配的数据支持画像。Enterprise SaaS and Big Tech / Platform contribute 51.7% of prior-employer sources, versus 8.4% from AI Lab / Model Co. and Academia / Research combined; together with a 1.34× Product / Design over-index and a 1.26× Research / Models over-index, the evidence supports prioritizing product and model builders from mature technology organizations.

02 · 人才结构02 · Talent structure

Engineering 占比接近，非 Engineering 配置把三家公司拉开Similar Engineering shares conceal sharply different non-Engineering structures

Engineering 只从 40.3% 到 44.7%；非工程侧却出现三个清晰重心：OpenAI 的 Product / Design 与 GTM / Customer 均为 12.7%，Anthropic 的 GTM / Customer 为 22.6%，xAI 的 Human Data & Evaluation 为 28.4%。Engineering ranges only from 40.3% to 44.7%, while the non-Engineering mix splits clearly: Product / Design and GTM / Customer are both 12.7% at OpenAI, GTM / Customer is 22.6% at Anthropic, and Human Data & Evaluation is 28.4% at xAI.

OpenAI · 岗位构成OpenAI · Role mix

OpenAI 没有单一非 Engineering 极点：Product / Design 与 GTM / Customer 均为 12.7%，Corporate Functions 为 10.9%；同时 Research / Models 占 6.8%，为三家公司最高。OpenAI has no single non-Engineering pole: Product / Design and GTM / Customer are both 12.7%, Corporate Functions is 10.9%, and Research / Models is the highest of the three companies at 6.8%.

Engineering

44.7%

Research / Models

6.8%

Applied AI

2.5%

Human Data & Evaluation

0.6%

Product / Design

12.7%

GTM / Customer

12.7%

People / Recruiting

6.4%

Corporate Functions

10.9%

Other / Unclassified

2.7%

数据来源: Metix AIData source: Metix AI

Anthropic · 岗位构成Anthropic · Role mix

Anthropic 的 GTM / Customer 占 22.6%，是 Research / Models 4.4% 的 5.1 倍，也分别比 OpenAI 和 xAI 高 9.9 与 17.0 个百分点；商业化配置是其最明显的组织差异。Anthropic’s GTM / Customer share is 22.6%, 5.1 times its 4.4% Research / Models share and respectively 9.9 and 17.0 percentage points above OpenAI and xAI; commercialization is its clearest organizational difference.

Engineering

41.6%

Research / Models

4.4%

Applied AI

2.5%

Human Data & Evaluation

0.2%

Product / Design

6.8%

GTM / Customer

22.6%

People / Recruiting

7.1%

Corporate Functions

12.7%

Other / Unclassified

2.3%

数据来源: Metix AIData source: Metix AI

xAI · 岗位构成xAI · Role mix

Human Data & Evaluation 占 28.4%，对应 425 条样本，而 Research / Models 仅占 2.1%；xAI 的辨识度来自大规模人类反馈、评估与数据运营。Human Data & Evaluation accounts for 28.4%, or 425 profiles, while Research / Models is only 2.1%; xAI is distinguished by scaled human-feedback, evaluation, and data operations.

Engineering

40.3%

Research / Models

2.1%

Applied AI

0.9%

Human Data & Evaluation

28.4%

Product / Design

2.9%

GTM / Customer

5.6%

People / Recruiting

3.3%

Corporate Functions

14.6%

Other / Unclassified

1.9%

数据来源: Metix AIData source: Metix AI

岗位族群 over-indexRole-family over-index

相对配置进一步否定了“同一种 AI 实验室”的假设：xAI 在 Human Data & Evaluation 上达到 5.83×，Anthropic 在 GTM / Customer 上为 1.58×，OpenAI 则在 Product / Design 与 Research / Models 上分别为 1.34× 和 1.26×。Relative allocation rejects the idea of a single AI-lab template: xAI reaches 5.83× in Human Data & Evaluation, Anthropic reaches 1.58× in GTM / Customer, and OpenAI reaches 1.34× in Product / Design and 1.26× in Research / Models.

OpenAI

Anthropic

xAI

Engineering

1.04×

0.96×

0.93×

Research / Models

1.26×

0.81×

0.40×

Applied AI

1.11×

1.12×

0.39×

Human Data & Evaluation

0.12×

0.03×

5.83×

Product / Design

1.34×

0.71×

0.30×

GTM / Customer

0.89×

1.58×

0.39×

People / Recruiting

1.05×

1.16×

0.55×

Corporate Functions

0.91×

1.06×

1.21×

Other / Unclassified

1.10×

0.93×

0.77×

数据来源: Metix AIData source: Metix AI

03 · 人才来源03 · Talent sources

成熟技术公司才是主要供给池，但三家公司吸收路径不同Mature technology companies are the main supply pool, but each company draws differently

OpenAI 与 Anthropic 的 Enterprise SaaS 和 Big Tech / Platform 来源合计分别为 51.7% 和 52.0%，但 OpenAI 更偏 Big Tech / Platform，Anthropic 更偏 Enterprise SaaS；xAI 两类合计仅 34.1%，来源更分散至 X、Academia / Research、硬件与咨询背景。Enterprise SaaS and Big Tech / Platform account for 51.7% of OpenAI sources and 52.0% of Anthropic sources, but OpenAI tilts toward Big Tech / Platform while Anthropic tilts toward Enterprise SaaS; the same two categories total only 34.1% at xAI, whose sources spread further across X, Academia / Research, hardware, and consulting.

OpenAI · 上一站公司类型OpenAI · Prior-employer types

Enterprise SaaS 与 Big Tech / Platform 合计占 51.7%，是 AI Lab / Model Co. 与 Academia / Research 合计 8.4% 的六倍以上；OpenAI 主要从成熟技术组织吸收人才。Enterprise SaaS and Big Tech / Platform account for 51.7%, more than six times the 8.4% combined share from AI Lab / Model Co. and Academia / Research; OpenAI draws mainly from mature technology organizations.

Enterprise SaaS

27.4%

Big Tech / Platform

24.3%

Finance / Consulting

9.8%

Auto / Hardware / Robotics

7.2%

Consumer Internet / Media

5.1%

Unknown Prior Company

5.0%

AI Lab / Model Co.

4.3%

Academia / Research

4.1%

Other categories

12.8%

数据来源: Metix AIData source: Metix AI

Anthropic · 上一站公司类型Anthropic · Prior-employer types

Enterprise SaaS 占 35.3%，超过 Big Tech / Platform 的两倍；结合 GTM / Customer 的 22.6%，Anthropic 当前的人才结构与来源共同呈现企业软件商业化导向。Enterprise SaaS reaches 35.3%, more than twice Big Tech / Platform; together with a 22.6% GTM / Customer share, Anthropic’s current talent mix and sources point to an enterprise-software commercialization orientation.

Enterprise SaaS

35.3%

Big Tech / Platform

16.7%

Finance / Consulting

10.0%

AI Lab / Model Co.

6.3%

Auto / Hardware / Robotics

5.9%

Academia / Research

4.7%

Consumer Internet / Media

4.3%

Other Business

3.6%

Other categories

13.2%

数据来源: Metix AIData source: Metix AI

xAI · 上一站公司类型xAI · Prior-employer types

最大具名来源类别 Enterprise SaaS 仅占 19.6%；Academia / Research 达 8.9%、Auto / Hardware / Robotics 达 9.4%，均高于另外两家公司。The largest named source category, Enterprise SaaS, is only 19.6%; Academia / Research reaches 8.9% and Auto / Hardware / Robotics reaches 9.4%, both above the other two companies.

Enterprise SaaS

19.6%

Big Tech / Platform

14.5%

Finance / Consulting

11.0%

Auto / Hardware / Robotics

9.4%

Academia / Research

8.9%

Other Business

6.6%

Consumer Internet / Media

5.5%

AI Lab / Model Co.

4.8%

Other categories

19.7%

数据来源: Metix AIData source: Metix AI

主要上一站公司Top prior employers

OpenAI · 主要上一站公司OpenAI · Top prior employers

Google、Meta 和 Apple 合计占 16.9%，前三大直接来源全部是大型平台公司；这与 51.7% 的成熟平台与 SaaS 类别占比相互印证。Google, Meta, and Apple account for 16.9% combined, and all three top direct sources are major platform companies, reinforcing the 51.7% mature-platform and SaaS category share.

Google

7.4%

Anthropic · 主要上一站公司Anthropic · Top prior employers

Google、Stripe 和 Meta 合计占 15.6%；Stripe 以 5.1% 接近 Google 的 6.4%，进一步确认 Enterprise SaaS 是 Anthropic 的核心供给池。Google, Stripe, and Meta account for 15.6% combined; Stripe at 5.1% is close to Google at 6.4%, reinforcing Enterprise SaaS as Anthropic’s core supply pool.

Google

6.4%

Stripe

5.1%

xAI · 主要上一站公司xAI · Top prior employers

X 以 5.1% 成为 xAI 最大单一上一站公司，高于 Google 的 2.3% 和 Microsoft 的 1.8%；创始人关联平台构成一条独有的直接来源通道。X is xAI’s largest single prior employer at 5.1%, above Google at 2.3% and Microsoft at 1.8%; the founder-linked platform forms a distinct direct-source channel.

5.1%

流动高度不对称，最清晰通道是 OpenAI → AnthropicMovement is highly asymmetric; the clearest corridor is OpenAI → Anthropic

OpenAI → Anthropic 有 47 条可见流动，是反向 10 条的 4.7 倍；其他任一方向均不超过 12 条，显示一条高度集中的相邻人才通道。OpenAI → Anthropic has 47 visible moves, 4.7 times the 10 in reverse; every other direction is 12 or fewer, indicating one highly concentrated adjacent talent corridor.

三家公司之间的可见直接流动Visible direct movement among the three companies

OpenAI → Anthropic 的 47 条记录高于其他任一方向，反向仅 10 条；这说明 OpenAI 是 Anthropic 最可见的相邻人才来源，但不能据此推断完整净流入。OpenAI → Anthropic reaches 47 visible moves, above every other direction, while the reverse path has only 10; OpenAI is Anthropic’s most visible adjacent talent source, but this does not establish complete net inflow.

OpenAI → Anthropic

xAI → OpenAI

Anthropic → OpenAI

OpenAI → xAI

xAI → Anthropic

Anthropic → xAI

数据来源: Metix AIData source: Metix AI

05 · 商业判断与建议动作05 · Commercial implications and recommended actions

同一套 AI 人才画像无法覆盖三家公司现有结构差异One AI talent profile cannot cover the three companies’ current structural differences

共同 Engineering 基础池可以复用，但画像应分别侧重 OpenAI 的产品与模型衔接、Anthropic 的企业软件商业化，以及 xAI 的评估与数据运营；来源池应随之拆分。The shared Engineering base can be reused, but profiles should emphasize product-model translation for OpenAI, enterprise-software commercialization for Anthropic, and evaluation and data operations for xAI; their source pools should diverge accordingly.

OpenAI

判断：Judgment: Product / Design 与 Research / Models 均呈 over-index，且 51.7% 的上一站来源来自成熟平台与 SaaS；核心画像是能把模型能力转成规模化产品的人。Product / Design and Research / Models both over-index, while 51.7% of prior-employer sources come from mature platforms and SaaS; the core profile is a builder who can turn model capability into scaled products.

动作：Action: 优先覆盖 Big Tech / Platform 与 Enterprise SaaS 中具备产品、平台或模型落地经历的资深建设者。Prioritize experienced builders in Big Tech / Platform and Enterprise SaaS with product, platform, or model-deployment experience.

Anthropic

判断：Judgment: GTM / Customer 占 22.6%，Enterprise SaaS 来源占 35.3%；Anthropic 的现有人才结构和来源更接近企业软件商业化团队；对应画像应优先覆盖企业软件商业化经验。GTM / Customer accounts for 22.6% and Enterprise SaaS for 35.3% of prior-employer sources; Anthropic’s current talent structure and sources align more closely with an enterprise-software commercialization team; the corresponding profile should prioritize enterprise-software commercialization experience.

动作：Action: 以 Enterprise SaaS 为第一来源池，优先识别解决方案、产品、销售工程与客户成功背景。Use Enterprise SaaS as the first source pool, prioritizing solutions, product, sales-engineering, and customer-success backgrounds.

xAI

判断：Judgment: Human Data & Evaluation 占 28.4%、over-index 为 5.83×；评估、数据质量与运营规模化是 xAI 最显著的非 Engineering 人才重心。Human Data & Evaluation accounts for 28.4% with a 5.83× over-index; Evaluation, data quality, and scaled operations form xAI’s most distinctive non-Engineering talent concentration.

动作：Action: 来源池同时覆盖 Academia / Research、Auto / Hardware / Robotics 与 X，重点寻找能把专业知识转成评估和数据流程的人。Cover Academia / Research, Auto / Hardware / Robotics, and X, focusing on people who can turn domain expertise into evaluation and data workflows.

组合建议：Portfolio recommendation: 共用 Engineering 基础池，但分别建立“产品与模型衔接”“企业软件商业化”“评估与数据运营”三套画像；公司来源只作为证据，不替代真实经历判断。Share the Engineering foundation, but maintain separate profiles for product-model translation, enterprise-software commercialization, and evaluation/data operations; prior employers are evidence, not a substitute for demonstrated experience.

专题 · 美国与重点国际市场Special analysis · U.S. vs priority international markets

xAI 的 Human Data & Evaluation 高集中度并非只出现在美国xAI’s Human Data & Evaluation concentration is not unique to the U.S.

在美国和 15 个重点国际市场中，xAI 的 Human Data & Evaluation 占比分别为 23.3% 和 34.4%，对最高同业的领先分别为 22.7 和 33.3 个百分点。OpenAI 与 Anthropic 在重点国际市场分别为 1.1% 和 0.7%。结果说明 xAI 的差异在两个地区组中都成立，但不能据此判断地理因素造成了差异。Human Data & Evaluation accounts for 23.3% of xAI profiles in the U.S. and 34.4% across 15 priority international markets, leading the highest peer by 22.7 and 33.3 percentage points, respectively. OpenAI and Anthropic stand at 1.1% and 0.7% in priority international markets. The difference holds in both geographic groups, but the comparison does not establish that geography caused it.

一致识别标准下，美国与重点国际市场的公司差异Company differences in the U.S. and priority international markets under one consistent definition

前文完整岗位构成显示 xAI 的 Human Data & Evaluation 占 28.4%、over-index 为 5.83×；按一致的跨地区识别标准复核后，美国为 23.3%、重点国际市场为 34.4%。两组指标的定义范围不同，不能相加，但都显示 xAI 显著高于同业。The full role mix places xAI’s Human Data & Evaluation share at 28.4% with a 5.83× over-index; under the consistent cross-market definition, the share is 23.3% in the U.S. and 34.4% in priority international markets. The measures cover different classification ranges and are not additive, but both place xAI well above its peers.

美国U.S.重点国际市场Priority international markets

OpenAI

美国U.S.

0.5% 28 / 5,313

国际市场International

1.1% 9 / 821

重点国际市场占比比美国高 0.6 个百分点，但绝对水平仍只有 1.1%。The priority international share is 0.6 percentage points above the U.S., but the absolute level remains only 1.1%.

Anthropic

美国U.S.

0.2% 6 / 2,612

国际市场International

0.7% 3 / 433

重点国际市场占比比美国高 0.5 个百分点，但仅有 3 条 Human Data & Evaluation 档案。The priority international share is 0.5 percentage points above the U.S., but it includes only 3 Human Data & Evaluation profiles.

xAI

美国U.S.

23.3% 348 / 1,496

国际市场International

34.4% 129 / 375

重点国际市场占比比美国高 11.1 个百分点；该差异不能用于判断地理因素的因果影响。The priority international share is 11.1 percentage points above the U.S.; the comparison does not establish a causal geography effect.

数据来源: Metix AIData source: Metix AI

差异不是由单一国家制造：xAI 的 Human Data & Evaluation 人才分布在多个国际市场The difference is not created by one country: xAI’s Human Data & Evaluation talent spans multiple international markets

xAI 的 129 条国际市场 Human Data & Evaluation 档案分布在 12 个非零市场。印度、加拿大、英国、日本和澳大利亚合计贡献 114 条，占 88.4%；其余 15 条分散在另外 7 个市场，因此不能把高占比归因于单一国家。xAI’s 129 international Human Data & Evaluation profiles span 12 non-zero markets. India, Canada, the United Kingdom, Japan, and Australia contribute 114, or 88.4%; the remaining 15 profiles are spread across 7 other markets, so the high share cannot be attributed to one country.

印度India

44 60.3%

加拿大Canada

30 49.2%

英国United Kingdom

18 18.4%

日本Japan

12 60.0%

澳大利亚Australia

10 62.5%

西班牙Spain

4 28.6%

韩国South Korea

3 60.0%

爱尔兰Ireland

2 6.5%

德国Germany

2 13.3%

比利时Belgium

2 100.0%

法国France

1 9.1%

瑞士Switzerland

1 33.3%

数据来源: Metix AIData source: Metix AI

两个地区组占比不同，排序一致Different regional shares, same ranking

xAI 对最高同业的领先在美国为 22.7 个百分点，在重点国际市场为 33.3 个百分点；两个地区组均由 xAI 居首，但不能据此判断地理因素造成了差异。xAI leads the highest peer by 22.7 percentage points in the U.S. and 33.3 percentage points in priority international markets; xAI ranks first in both geographic groups, but the comparison does not establish that geography caused the difference.

印度是共同市场，但 Human Data & Evaluation 配置不同India is a shared market, but Human Data & Evaluation configurations differ

三家公司在印度都有 Human Data & Evaluation 档案，但 xAI 为 44 / 73，OpenAI 为 4 / 109，Anthropic 为 1 / 40；同一市场中的配置差异仍然成立。All three companies have Human Data & Evaluation profiles in India, but xAI has 44 of 73 profiles, versus 4 of 109 at OpenAI and 1 of 40 at Anthropic; the configuration difference still holds within the same market.

xAI 应单列 Human Data & Evaluation 画像xAI warrants a separate Human Data & Evaluation profile

完整岗位分类与跨地区统一指标的定义范围不同，不能相加，但都显示 xAI 显著高于同业。竞品人才地图不应把这组人并入 Research / Models，否则会遗漏 xAI 最具辨识度的人才配置。The full role classification and the consistent cross-market measure cover different ranges and are not additive, but both place xAI well above its peers. Competitive talent maps should not fold this population into Research / Models, or they will miss xAI’s most distinctive talent configuration.

06 · 范围与使用边界06 · Scope and limitations

结论Conclusion

岗位、来源与流动对比覆盖 9,421 条美国公开职业档案；地区对比还纳入 15 个重点国际市场的 1,629 条档案。所有指标均用于相对比较，不用于推算全球员工总量。The role, source, and movement comparison covers 9,421 U.S. public professional profiles; the geographic comparison also includes 1,629 profiles across 15 priority international markets. All measures support relative comparison and are not used to estimate global employee totals.

分析范围Analysis scope

美国组用于岗位构成、上一站来源和公司间流动；上一站公司识别覆盖率为 OpenAI 95.0%、Anthropic 98.0%、xAI 97.8%。The U.S. group supports role mix, prior-employer sources, and inter-company movement; prior-employer coverage is 95.0% for OpenAI, 98.0% for Anthropic, and 97.8% for xAI.
地区模块覆盖英国、爱尔兰、日本、新加坡、印度、德国、法国、瑞士、韩国、澳大利亚、加拿大、西班牙、瑞典、比利时和巴西，不代表完整非美国市场。The geographic module covers the United Kingdom, Ireland, Japan, Singapore, India, Germany, France, Switzerland, South Korea, Australia, Canada, Spain, Sweden, Belgium, and Brazil; it is not complete non-U.S. coverage.
地区比较采用统一的 Human Data & Evaluation 识别标准；该指标比前文完整岗位分类更窄，不能与 28.4% 直接相加。The geographic comparison applies one consistent Human Data & Evaluation definition; this measure is narrower than the full role classification above and cannot be added to the 28.4% share.

主要限制Key limitations

公开职业档案可能存在更新滞后、自我呈现偏差和岗位命名差异。Public profiles may contain update lags, self-presentation bias, and inconsistent titles.
国家级小样本占比波动较大，应同时查看绝对人数；比利时、瑞士等低基数市场不单独支撑总体判断。Country-level shares can be volatile at small sample sizes and should be read with absolute counts; low-base markets such as Belgium and Switzerland do not support the overall conclusion alone.
公司间流动不代表完整流入、流出或留存。Inter-company movement does not represent complete inflow, outflow, or retention.

想要 OpenAI、Anthropic 或 xAI 的完整名单，或换成你的目标公司？Want the full list for OpenAI, Anthropic, or xAI—or your own target company?

岗位结构、人才来源、公司间流动与地区分布，都可按需为任意一家 AI 公司单独生成，并对接可联系的候选人。Role mix, talent sources, inter-company movement, and geography can be generated on demand for any AI company and connected to reachable candidates.

人才报告 · 不展示个人信息 · 由 Metix AI 提供Talent report · no personal information shown · provided by Metix AI