Tech Sep 8, 2026

OpenAI, Anthropic, Google DeepMind: Comparing Talent Structures

From research and engineering to product and commercial roles: comparing the people, experience, education and pay behind three frontier AI companies.

THREE COMPANIES, THREE TALENT STRUCTURES

Updated

Published profile studies show a larger product / design share at OpenAI, a stronger commercial and customer presence at Anthropic, and more explicit research roles in Google DeepMind’s technical titles.

OpenAI

Product / Design has a larger share in the US sample; MTS dominates the technical titles.

Company boundary

M1 / M2 use report-identified company profiles; legal-entity and contractor treatment are not fully disclosed. These are not official Foundation or global operating-company rosters.

Anthropic

Commercial and customer roles (GTM / Customer) have a larger share in the same US sample.

Company boundary

Report-identified Anthropic profiles are not a global Anthropic PBC census. Official security-adjacent disclosure is a separate historical, cross-functional measure.

Google DeepMind

Technical profiles show a clearer distinction between research scientist and research engineer titles.

Company boundary

G1 identifies the combined DeepMind and Brain unit. G2 states that DeepMind Technologies Limited’s workforce is employed by other Alphabet companies. Team, legal entity and parent company are distinct.

01

Functions: the clearest differences are in product and commercial roles

Engineering accounts for over 40% in both samples. Product / design is 12.7% at OpenAI and 6.8% at Anthropic; GTM / Customer is 12.7% and 22.6%. Select a function for details.

Metix M1 · Published Jul 2026 · US current-employment profiles: OpenAI n=5,313; Anthropic n=2,612. Google DeepMind is outside this report’s scope.

Study scope

Limited evidence | Metix M1, published 2026-07-23, modified 2026-08-20; exact snapshot undisclosed. US current-employment profiles only: OpenAI n=5,313; Anthropic n=2,612. GDM not covered. Shares describe source labels; MTS classification and cell counts are undisclosed.

Three-company comparison of nine functions and source-defined combined categories; full evidence for the selected function follows
FunctionOpenAIAnthropicGoogle DeepMind
OpenAI
6.8%
Anthropic
4.4%
Google DeepMind
Outside report scope
OpenAI
44.7%
Anthropic
41.6%
Google DeepMind
Outside report scope
OpenAI
12.7%
Anthropic
6.8%
Google DeepMind
Outside report scope
OpenAI
Not separately disclosed
Anthropic
Not separately disclosed
Google DeepMind
Outside report scope
OpenAI
Not separately disclosed
Anthropic
Not separately disclosed
Google DeepMind
Outside report scope
OpenAI
Not separately disclosed
Anthropic
Not separately disclosed
Google DeepMind
Outside report scope
OpenAI
6.4%
Anthropic
7.1%
Google DeepMind
Outside report scope
OpenAI
Not separately disclosed
Anthropic
Not separately disclosed
Google DeepMind
Outside report scope
OpenAI
Not separately disclosed
Anthropic
Not separately disclosed
Google DeepMind
Outside report scope
OpenAI
12.7%
Anthropic
22.6%
Google DeepMind
Outside report scope
OpenAI
10.9%
Anthropic
12.7%
Google DeepMind
Outside report scope
Function details & sources

Research / ModelsEvidence by company

OpenAI

Research / Models (source label)

6.8%

Metix AI · M1

Scope & source details
Company entity
OpenAI
Count / share / amount
6.8%
Denominator
Company US current-employment profiles n=5313, not global staff
Sample size
5313
As of
2026-07-23 report; snapshot not disclosed / 快照未披露
Evidence type
Institutional sample, not official staffing; classification not independently validated
Population / scope
US, multi-function profiles; original classification
Gaps / limitations
MTS assignment undisclosed. Compare report labels only, not complete research or engineering teams. Counts unpublished; not back-calculated.
Evidence locator
Section 02, company Role mix
Observation ID
function-research-openai

Anthropic

Research / Models (source label)

4.4%

Metix AI · M1

Scope & source details
Company entity
Anthropic
Count / share / amount
4.4%
Denominator
Company US current-employment profiles n=2612, not global staff
Sample size
2612
As of
2026-07-23 report; snapshot not disclosed / 快照未披露
Evidence type
Institutional sample, not official staffing; classification not independently validated
Population / scope
US, multi-function profiles; original classification
Gaps / limitations
MTS assignment undisclosed. Compare report labels only, not complete research or engineering teams. Counts unpublished; not back-calculated.
Evidence locator
Section 02, company Role mix
Observation ID
function-research-anthropic

Google DeepMind

Research / Models (source label)

Outside report scope

Metix AI · M1 / SignalFire · S1 / Google DeepMind · G1 / DeepMind Technologies Limited / Companies House · G2

Scope & source details
Company entity
Google DeepMind
Count / share / amount
Outside report scope
Denominator
Not applicable: Google DeepMind is not included in this report
Sample size
Not applicable: outside source scope
As of
2026-07-23 report; snapshot not disclosed / 快照未披露
Evidence type
Institutional sample, not official staffing; classification not independently validated
Population / scope
Google DeepMind outside M1 scope
Gaps / limitations
M1’s third company is xAI; S1 only gives a six-lab aggregate. US/UK/France technical samples or parent Google data cannot fill this gap.
Evidence locator
M1 section 06; S1 methodology; G1; G2 printed p.3
Observation ID
function-research-deepmind
02

Technical teams: three different title systems

Google DeepMind has more explicit research scientist and research engineer titles. OpenAI and Anthropic use MTS (Member of Technical Staff) extensively, shown separately below.

Metix M2 · Approx. H1 2026 · US/UK/France technical profiles; n = 2,628 / 1,251 / 2,837 respectively. Shares use each technical sample as the denominator.

Study scope

Limited evidence | Metix M2, approximately H1 2026, US/UK/France technical profiles, n=2,628 / 1,251 / 2,837. Counts from source chart; shares = count / same-source technical pool. Neither all-staff shares nor a research/engineering capability ranking.

Research ScientistOpenAI210 people · 8.0%Anthropic68 people · 5.4%Google DeepMind918 people · 32.4%View details

OpenAI

Research Scientist

210 people · 8.0%

Metix AI · M2

Scope & source details
Company entity
OpenAI
Count / share / amount
210 people · 8.0%
Denominator
Same-report company technical pool n=2628, not all employees
Sample size
2628
As of
Approx. H1 2026 / 约 2026 年上半年
Evidence type
Institutional sample, not official staffing; classification not independently validated
Population / scope
US/UK/France technical profiles; title/headline labels
Gaps / limitations
Counts transcribed from original SVG; shares calculated here with the same-source technical denominator. MTS stays unspecified; Safety is not added again to research/engineering; Google affiliations may be missed for DeepMind.
Evidence locator
Section 3.4 Function × lab matrix; 9.1 methodology
Observation ID
technical-research-scientist-openai

Anthropic

Research Scientist

68 people · 5.4%

Metix AI · M2

Scope & source details
Company entity
Anthropic
Count / share / amount
68 people · 5.4%
Denominator
Same-report company technical pool n=1251, not all employees
Sample size
1251
As of
Approx. H1 2026 / 约 2026 年上半年
Evidence type
Institutional sample, not official staffing; classification not independently validated
Population / scope
US/UK/France technical profiles; title/headline labels
Gaps / limitations
Counts transcribed from original SVG; shares calculated here with the same-source technical denominator. MTS stays unspecified; Safety is not added again to research/engineering; Google affiliations may be missed for DeepMind.
Evidence locator
Section 3.4 Function × lab matrix; 9.1 methodology
Observation ID
technical-research-scientist-anthropic

Google DeepMind

Research Scientist

918 people · 32.4%

Metix AI · M2

Scope & source details
Company entity
Google DeepMind
Count / share / amount
918 people · 32.4%
Denominator
Same-report company technical pool n=2837, not all employees
Sample size
2837
As of
Approx. H1 2026 / 约 2026 年上半年
Evidence type
Institutional sample, not official staffing; classification not independently validated
Population / scope
US/UK/France technical profiles; title/headline labels
Gaps / limitations
Counts transcribed from original SVG; shares calculated here with the same-source technical denominator. MTS stays unspecified; Safety is not added again to research/engineering; Google affiliations may be missed for DeepMind.
Evidence locator
Section 3.4 Function × lab matrix; 9.1 methodology
Observation ID
technical-research-scientist-deepmind
Research EngineerOpenAI43 people · 1.6%Anthropic12 people · 1.0%Google DeepMind507 people · 17.9%View details

OpenAI

Research Engineer

43 people · 1.6%

Metix AI · M2

Scope & source details
Company entity
OpenAI
Count / share / amount
43 people · 1.6%
Denominator
Same-report company technical pool n=2628, not all employees
Sample size
2628
As of
Approx. H1 2026 / 约 2026 年上半年
Evidence type
Institutional sample, not official staffing; classification not independently validated
Population / scope
US/UK/France technical profiles; title/headline labels
Gaps / limitations
Counts transcribed from original SVG; shares calculated here with the same-source technical denominator. MTS stays unspecified; Safety is not added again to research/engineering; Google affiliations may be missed for DeepMind.
Evidence locator
Section 3.4 Function × lab matrix; 9.1 methodology
Observation ID
technical-research-engineer-openai

Anthropic

Research Engineer

12 people · 1.0%

Metix AI · M2

Scope & source details
Company entity
Anthropic
Count / share / amount
12 people · 1.0%
Denominator
Same-report company technical pool n=1251, not all employees
Sample size
1251
As of
Approx. H1 2026 / 约 2026 年上半年
Evidence type
Institutional sample, not official staffing; classification not independently validated
Population / scope
US/UK/France technical profiles; title/headline labels
Gaps / limitations
Counts transcribed from original SVG; shares calculated here with the same-source technical denominator. MTS stays unspecified; Safety is not added again to research/engineering; Google affiliations may be missed for DeepMind.
Evidence locator
Section 3.4 Function × lab matrix; 9.1 methodology
Observation ID
technical-research-engineer-anthropic

Google DeepMind

Research Engineer

507 people · 17.9%

Metix AI · M2

Scope & source details
Company entity
Google DeepMind
Count / share / amount
507 people · 17.9%
Denominator
Same-report company technical pool n=2837, not all employees
Sample size
2837
As of
Approx. H1 2026 / 约 2026 年上半年
Evidence type
Institutional sample, not official staffing; classification not independently validated
Population / scope
US/UK/France technical profiles; title/headline labels
Gaps / limitations
Counts transcribed from original SVG; shares calculated here with the same-source technical denominator. MTS stays unspecified; Safety is not added again to research/engineering; Google affiliations may be missed for DeepMind.
Evidence locator
Section 3.4 Function × lab matrix; 9.1 methodology
Observation ID
technical-research-engineer-deepmind
MTS · unspecifiedOpenAI1,702 people · 64.8%Anthropic962 people · 76.9%Google DeepMind3 people · 0.1%View details

OpenAI

MTS · unspecified

1,702 people · 64.8%

Metix AI · M2

Scope & source details
Company entity
OpenAI
Count / share / amount
1,702 people · 64.8%
Denominator
Same-report company technical pool n=2628, not all employees
Sample size
2628
As of
Approx. H1 2026 / 约 2026 年上半年
Evidence type
Institutional sample, not official staffing; classification not independently validated
Population / scope
US/UK/France technical profiles; title/headline labels
Gaps / limitations
Counts transcribed from original SVG; shares calculated here with the same-source technical denominator. MTS stays unspecified; Safety is not added again to research/engineering; Google affiliations may be missed for DeepMind.
Evidence locator
Section 3.4 Function × lab matrix; 9.1 methodology
Observation ID
technical-mts-openai

Anthropic

MTS · unspecified

962 people · 76.9%

Metix AI · M2

Scope & source details
Company entity
Anthropic
Count / share / amount
962 people · 76.9%
Denominator
Same-report company technical pool n=1251, not all employees
Sample size
1251
As of
Approx. H1 2026 / 约 2026 年上半年
Evidence type
Institutional sample, not official staffing; classification not independently validated
Population / scope
US/UK/France technical profiles; title/headline labels
Gaps / limitations
Counts transcribed from original SVG; shares calculated here with the same-source technical denominator. MTS stays unspecified; Safety is not added again to research/engineering; Google affiliations may be missed for DeepMind.
Evidence locator
Section 3.4 Function × lab matrix; 9.1 methodology
Observation ID
technical-mts-anthropic

Google DeepMind

MTS · unspecified

3 people · 0.1%

Metix AI · M2

Scope & source details
Company entity
Google DeepMind
Count / share / amount
3 people · 0.1%
Denominator
Same-report company technical pool n=2837, not all employees
Sample size
2837
As of
Approx. H1 2026 / 约 2026 年上半年
Evidence type
Institutional sample, not official staffing; classification not independently validated
Population / scope
US/UK/France technical profiles; title/headline labels
Gaps / limitations
Counts transcribed from original SVG; shares calculated here with the same-source technical denominator. MTS stays unspecified; Safety is not added again to research/engineering; Google affiliations may be missed for DeepMind.
Evidence locator
Section 3.4 Function × lab matrix; 9.1 methodology
Observation ID
technical-mts-deepmind
Engineering · broad categoryOpenAI607 people · 23.1%Anthropic164 people · 13.1%Google DeepMind1,388 people · 48.9%View details

OpenAI

Engineering · broad category

607 people · 23.1%

Metix AI · M2

Scope & source details
Company entity
OpenAI
Count / share / amount
607 people · 23.1%
Denominator
Same-report company technical pool n=2628, not all employees
Sample size
2628
As of
Approx. H1 2026 / 约 2026 年上半年
Evidence type
Institutional sample, not official staffing; classification not independently validated
Population / scope
US/UK/France technical profiles; title/headline labels
Gaps / limitations
Counts transcribed from original SVG; shares calculated here with the same-source technical denominator. MTS stays unspecified; Safety is not added again to research/engineering; Google affiliations may be missed for DeepMind.
Evidence locator
Section 3.4 Function × lab matrix; 9.1 methodology
Observation ID
technical-engineering-broad-openai

Anthropic

Engineering · broad category

164 people · 13.1%

Metix AI · M2

Scope & source details
Company entity
Anthropic
Count / share / amount
164 people · 13.1%
Denominator
Same-report company technical pool n=1251, not all employees
Sample size
1251
As of
Approx. H1 2026 / 约 2026 年上半年
Evidence type
Institutional sample, not official staffing; classification not independently validated
Population / scope
US/UK/France technical profiles; title/headline labels
Gaps / limitations
Counts transcribed from original SVG; shares calculated here with the same-source technical denominator. MTS stays unspecified; Safety is not added again to research/engineering; Google affiliations may be missed for DeepMind.
Evidence locator
Section 3.4 Function × lab matrix; 9.1 methodology
Observation ID
technical-engineering-broad-anthropic

Google DeepMind

Engineering · broad category

1,388 people · 48.9%

Metix AI · M2

Scope & source details
Company entity
Google DeepMind
Count / share / amount
1,388 people · 48.9%
Denominator
Same-report company technical pool n=2837, not all employees
Sample size
2837
As of
Approx. H1 2026 / 约 2026 年上半年
Evidence type
Institutional sample, not official staffing; classification not independently validated
Population / scope
US/UK/France technical profiles; title/headline labels
Gaps / limitations
Counts transcribed from original SVG; shares calculated here with the same-source technical denominator. MTS stays unspecified; Safety is not added again to research/engineering; Google affiliations may be missed for DeepMind.
Evidence locator
Section 3.4 Function × lab matrix; 9.1 methodology
Observation ID
technical-engineering-broad-deepmind
Safety & Alignment · source categoryOpenAI66 people · 2.5%Anthropic45 people · 3.6%Google DeepMind21 people · 0.7%View details

OpenAI

Safety & Alignment · source category

66 people · 2.5%

Metix AI · M2

Scope & source details
Company entity
OpenAI
Count / share / amount
66 people · 2.5%
Denominator
Same-report company technical pool n=2628, not all employees
Sample size
2628
As of
Approx. H1 2026 / 约 2026 年上半年
Evidence type
Institutional sample, not official staffing; classification not independently validated
Population / scope
US/UK/France technical profiles; title/headline labels
Gaps / limitations
Counts transcribed from original SVG; shares calculated here with the same-source technical denominator. MTS stays unspecified; Safety is not added again to research/engineering; Google affiliations may be missed for DeepMind.
Evidence locator
Section 3.4 Function × lab matrix; 9.1 methodology
Observation ID
technical-safety-alignment-openai

Anthropic

Safety & Alignment · source category

45 people · 3.6%

Metix AI · M2

Scope & source details
Company entity
Anthropic
Count / share / amount
45 people · 3.6%
Denominator
Same-report company technical pool n=1251, not all employees
Sample size
1251
As of
Approx. H1 2026 / 约 2026 年上半年
Evidence type
Institutional sample, not official staffing; classification not independently validated
Population / scope
US/UK/France technical profiles; title/headline labels
Gaps / limitations
Counts transcribed from original SVG; shares calculated here with the same-source technical denominator. MTS stays unspecified; Safety is not added again to research/engineering; Google affiliations may be missed for DeepMind.
Evidence locator
Section 3.4 Function × lab matrix; 9.1 methodology
Observation ID
technical-safety-alignment-anthropic

Google DeepMind

Safety & Alignment · source category

21 people · 0.7%

Metix AI · M2

Scope & source details
Company entity
Google DeepMind
Count / share / amount
21 people · 0.7%
Denominator
Same-report company technical pool n=2837, not all employees
Sample size
2837
As of
Approx. H1 2026 / 约 2026 年上半年
Evidence type
Institutional sample, not official staffing; classification not independently validated
Population / scope
US/UK/France technical profiles; title/headline labels
Gaps / limitations
Counts transcribed from original SVG; shares calculated here with the same-source technical denominator. MTS stays unspecified; Safety is not added again to research/engineering; Google affiliations may be missed for DeepMind.
Evidence locator
Section 3.4 Function × lab matrix; 9.1 methodology
Observation ID
technical-safety-alignment-deepmind
Software engineering (narrow)OpenAINot separately disclosedAnthropicNot separately disclosedGoogle DeepMindNot separately disclosedView details

OpenAI

Software engineering (narrow)

Not separately disclosed

Metix AI · M2

Scope & source details
Company entity
OpenAI
Count / share / amount
Not separately disclosed
Denominator
Not disclosed by source
Sample size
2628
As of
Approx. H1 2026 / 约 2026 年上半年
Evidence type
Institutional sample, not official staffing; classification not independently validated
Population / scope
M2 technical classification
Gaps / limitations
Engineering is not decomposed into this subcategory; the broad group cannot be renamed software or platform engineering.
Evidence locator
Section 3.4
Observation ID
technical-software-engineering-openai

Anthropic

Software engineering (narrow)

Not separately disclosed

Metix AI · M2

Scope & source details
Company entity
Anthropic
Count / share / amount
Not separately disclosed
Denominator
Not disclosed by source
Sample size
1251
As of
Approx. H1 2026 / 约 2026 年上半年
Evidence type
Institutional sample, not official staffing; classification not independently validated
Population / scope
M2 technical classification
Gaps / limitations
Engineering is not decomposed into this subcategory; the broad group cannot be renamed software or platform engineering.
Evidence locator
Section 3.4
Observation ID
technical-software-engineering-anthropic

Google DeepMind

Software engineering (narrow)

Not separately disclosed

Metix AI · M2

Scope & source details
Company entity
Google DeepMind
Count / share / amount
Not separately disclosed
Denominator
Not disclosed by source
Sample size
2837
As of
Approx. H1 2026 / 约 2026 年上半年
Evidence type
Institutional sample, not official staffing; classification not independently validated
Population / scope
M2 technical classification
Gaps / limitations
Engineering is not decomposed into this subcategory; the broad group cannot be renamed software or platform engineering.
Evidence locator
Section 3.4
Observation ID
technical-software-engineering-deepmind
Infrastructure / platformOpenAINot separately disclosedAnthropicNot separately disclosedGoogle DeepMindNot separately disclosedView details

OpenAI

Infrastructure / platform

Not separately disclosed

Metix AI · M2

Scope & source details
Company entity
OpenAI
Count / share / amount
Not separately disclosed
Denominator
Not disclosed by source
Sample size
2628
As of
Approx. H1 2026 / 约 2026 年上半年
Evidence type
Institutional sample, not official staffing; classification not independently validated
Population / scope
M2 technical classification
Gaps / limitations
Engineering is not decomposed into this subcategory; the broad group cannot be renamed software or platform engineering.
Evidence locator
Section 3.4
Observation ID
technical-infrastructure-platform-openai

Anthropic

Infrastructure / platform

Not separately disclosed

Metix AI · M2

Scope & source details
Company entity
Anthropic
Count / share / amount
Not separately disclosed
Denominator
Not disclosed by source
Sample size
1251
As of
Approx. H1 2026 / 约 2026 年上半年
Evidence type
Institutional sample, not official staffing; classification not independently validated
Population / scope
M2 technical classification
Gaps / limitations
Engineering is not decomposed into this subcategory; the broad group cannot be renamed software or platform engineering.
Evidence locator
Section 3.4
Observation ID
technical-infrastructure-platform-anthropic

Google DeepMind

Infrastructure / platform

Not separately disclosed

Metix AI · M2

Scope & source details
Company entity
Google DeepMind
Count / share / amount
Not separately disclosed
Denominator
Not disclosed by source
Sample size
2837
As of
Approx. H1 2026 / 约 2026 年上半年
Evidence type
Institutional sample, not official staffing; classification not independently validated
Population / scope
M2 technical classification
Gaps / limitations
Engineering is not decomposed into this subcategory; the broad group cannot be renamed software or platform engineering.
Evidence locator
Section 3.4
Observation ID
technical-infrastructure-platform-deepmind
Historical context: security-adjacent (cross-functional)OpenAIOutside this disclosureAnthropicAround 8% of employeesGoogle DeepMindOutside this disclosureAnthropic official disclosure · 20 May 2024 · Cross-functional share of staff at that timeView details

OpenAI

Historical context: security-adjacent (cross-functional)

Outside this disclosure

Anthropic · A1

Scope & source details
Company entity
OpenAI
Count / share / amount
Outside this disclosure
Denominator
Not applicable: A1 only discloses Anthropic
Sample size
Not applicable: outside source scope
As of
2024-05-20
Evidence type
This official source does not cover this company
Population / scope
A1 is an Anthropic-only historical disclosure
Gaps / limitations
Not equivalent to AI safety, policy or legal. Not additive to the technical sample. This official disclosure only covers Anthropic.
Evidence locator
Paragraph on scaling up the security program
Observation ID
technical-security-adjacent-openai

Anthropic

Historical context: security-adjacent (cross-functional)

Around 8% of employees

Anthropic · A1

Scope & source details
Company entity
Anthropic
Count / share / amount
Around 8% of employees
Denominator
All Anthropic employees then; headcount not disclosed
Sample size
Source does not disclose valid n for this metric
As of
2024-05-20
Evidence type
Official company disclosure
Population / scope
Historical; original security-adjacent definition
Gaps / limitations
Not equivalent to AI safety, policy or legal. Not additive to the technical sample. This official disclosure only covers Anthropic.
Evidence locator
Paragraph on scaling up the security program
Observation ID
technical-security-adjacent-anthropic

Google DeepMind

Historical context: security-adjacent (cross-functional)

Outside this disclosure

Anthropic · A1

Scope & source details
Company entity
Google DeepMind
Count / share / amount
Outside this disclosure
Denominator
Not applicable: A1 only discloses Anthropic
Sample size
Not applicable: outside source scope
As of
2024-05-20
Evidence type
This official source does not cover this company
Population / scope
A1 is an Anthropic-only historical disclosure
Gaps / limitations
Not equivalent to AI safety, policy or legal. Not additive to the technical sample. This official disclosure only covers Anthropic.
Evidence locator
Paragraph on scaling up the security program
Observation ID
technical-security-adjacent-deepmind
Scale context

Metix M2 · Approx. H1 2026 · Visible employment profiles in the US, UK and France. Gaps in global headcount data are shown separately below.

OpenAI

Global company headcount

Not obtained in this review

Metix AI · M1 / Metix AI · M2

Scope & source details
Company entity
OpenAI
Count / share / amount
Not obtained in this review
Denominator
Usable denominator not obtained in this review
Sample size
Usable sample size not obtained in this review
As of
2026-09-08
Evidence type
Institutional sample, not official staffing; classification not independently validated
Population / scope
Target is the global organization; qualifying global headcount not obtained
Gaps / limitations
Visible profiles, technical pools, UK legal-entity counts and parent-company staff cannot substitute for the global team.
Evidence locator
See source scopes and exclusion register
Observation ID
scale-global-headcount-openai

Scale reference: visible US/UK/France profiles

5,495 people

Metix AI · M2

Scope & source details
Company entity
OpenAI
Count / share / amount
5,495 people
Denominator
Report-visible current-employment profile set; count, not a ratio
Sample size
5495
As of
Approx. H1 2026 / 约 2026 年上半年
Evidence type
Institutional sample, not official staffing; classification not independently validated
Population / scope
US/UK/France; multi-function profiles; not official headcount
Gaps / limitations
Explains sample coverage only; not company headcount and not multiplied by another report’s shares.
Evidence locator
Appendix 9.1, 9.2
Observation ID
scale-visible-profiles-openai

Anthropic

Global company headcount

Not obtained in this review

Metix AI · M1 / Metix AI · M2

Scope & source details
Company entity
Anthropic
Count / share / amount
Not obtained in this review
Denominator
Usable denominator not obtained in this review
Sample size
Usable sample size not obtained in this review
As of
2026-09-08
Evidence type
Institutional sample, not official staffing; classification not independently validated
Population / scope
Target is the global organization; qualifying global headcount not obtained
Gaps / limitations
Visible profiles, technical pools, UK legal-entity counts and parent-company staff cannot substitute for the global team.
Evidence locator
See source scopes and exclusion register
Observation ID
scale-global-headcount-anthropic

Scale reference: visible US/UK/France profiles

2,740 people

Metix AI · M2

Scope & source details
Company entity
Anthropic
Count / share / amount
2,740 people
Denominator
Report-visible current-employment profile set; count, not a ratio
Sample size
2740
As of
Approx. H1 2026 / 约 2026 年上半年
Evidence type
Institutional sample, not official staffing; classification not independently validated
Population / scope
US/UK/France; multi-function profiles; not official headcount
Gaps / limitations
Explains sample coverage only; not company headcount and not multiplied by another report’s shares.
Evidence locator
Appendix 9.1, 9.2
Observation ID
scale-visible-profiles-anthropic

Google DeepMind

Global company headcount

Not obtained in this review

Google DeepMind · G1 / DeepMind Technologies Limited / Companies House · G2

Scope & source details
Company entity
Google DeepMind
Count / share / amount
Not obtained in this review
Denominator
Usable denominator not obtained in this review
Sample size
Usable sample size not obtained in this review
As of
2026-09-08
Evidence type
Institutional sample, not official staffing; classification not independently validated
Population / scope
Target is the global organization; qualifying global headcount not obtained
Gaps / limitations
Visible profiles, technical pools, UK legal-entity counts and parent-company staff cannot substitute for the global team.
Evidence locator
See source scopes and exclusion register
Observation ID
scale-global-headcount-deepmind

Scale reference: visible US/UK/France profiles

4,123 people

Metix AI · M2

Scope & source details
Company entity
Google DeepMind
Count / share / amount
4,123 people
Denominator
Report-visible current-employment profile set; count, not a ratio
Sample size
4123
As of
Approx. H1 2026 / 约 2026 年上半年
Evidence type
Institutional sample, not official staffing; classification not independently validated
Population / scope
US/UK/France; multi-function profiles; not official headcount
Gaps / limitations
Explains sample coverage only; not company headcount and not multiplied by another report’s shares.
Evidence locator
Appendix 9.1, 9.2
Observation ID
scale-visible-profiles-deepmind

Save this comparison

Download the interactive article or structured data, with full sources.

All three companies build frontier models, but published talent samples reveal different organizational emphases: product and design stand out at OpenAI, commercial and customer roles at Anthropic, and explicit research titles at Google DeepMind. This article brings together institutional reports on functions, technical roles and career backgrounds to examine how those differences connect to the work each organization does. The function comparison currently covers US samples for the first two companies; the technical comparison covers all three in the US, UK and France. Detailed definitions and evidence accompany the interactive charts and appendix.

The clearest difference appears at the product and customer ends of the organization. In the same Metix US sample, Product / Design accounts for 12.7% at OpenAI and 6.8% at Anthropic; GTM / Customer accounts for 12.7% and 22.6%, respectively. The first category concerns how people use a product, while the second connects commercial development with customer adoption. This offers one way to read their organizational priorities: building the product experience is more prominent in OpenAI’s sample, while bringing the product into customers’ businesses is more prominent in Anthropic’s. Original report, sections 02 and 06

Hiring backgrounds add texture to that picture. The immediate-prior-employer categories show a stronger enterprise-software presence at Anthropic and a stronger large-technology-platform presence at OpenAI. These patterns align with the functional differences: building products for users and connecting models to enterprise work call for different combinations of experience. The background view therefore separates immediate previous employers, employers appearing anywhere in a career, and career-record spans, allowing readers to follow the functional comparison into the experience people bring.

Within technical teams, job titles become a revealing difference of their own. OpenAI and Anthropic make extensive use of Member of Technical Staff (MTS) in the samples, while Google DeepMind more often explicitly distinguishes Research Scientists from Research Engineers. More of the first two companies’ division of work sits beneath a shared title; DeepMind’s research track is easier to read from titles. The interactive comparison therefore gives MTS its own column alongside research, engineering and Safety & Alignment. Recorded PhDs provide an additional view of technical-team backgrounds. Original report, sections 3.4 and 9.1

Google DeepMind also has a distinctive organizational setting. It was formed by combining DeepMind and Google Research’s Brain team in 2023, placing the research organization within the larger Google and Alphabet system. Official organizational announcement The 2024 financial statements of the UK entity DeepMind Technologies Limited say that its workforce is directly employed by other Alphabet group companies. Understanding the organization means considering both the research team and the division of work across the group. The comparison consequently preserves Google DeepMind’s team boundary and marks its missing full-function breakdown in the third column. Official financial statements, printed pages 3 and 20

Pay adds a cost perspective to the technical-team comparison. BCG provides a 2024 annual MTS salary reference of US$353,000 for OpenAI and US$320,000 for Anthropic. These historical references draw on H-1B-related position data, exclude equity and bonuses, and do not cover Google DeepMind. Reading them after the role breakdown is more informative than looking at company salary figures alone: the particular role, seniority and location still determine which technical positions make a useful comparison. BCG, PDF pages 7–8, Exhibit 5.2

The most useful additions would be Google DeepMind’s full function distribution, the division of work within the combined commercial/customer category, and the research, software-engineering and infrastructure work performed under MTS titles. They would connect the product, commercial and research-track differences already visible to more specific teams.

Download complete JSON dataset · Download observations CSV · Download source register

COMMENTS

LEAVE A COMMENT

Comments are public and posted under your chosen name. The site owner may remove anything off-topic or abusive.

← Back to blog