Google DeepMind published AlphaFold 3 in Nature. Its researchers author the papers that define the agent era. Gemini Ultra runs on infrastructure DeepMind built. In August 2026, the organization scored 60 out of 100 on the ENTRA Pay Transparency Index, placing last among 20 scored AI employers.
The score is not a verdict on what DeepMind pays. Mountain View Senior Research Scientists clear $175,000 to $280,000 in base salary (Levels.fyi Q2 2026). London Research Scientists draw £130,000 to £220,000. The organization competes for, and attracts, the most sought-after AI researchers on the planet. The score is a verdict on what DeepMind publishes before a candidate applies, and those are two different things.
Salary Transparency Month August 2026 opened with Hugging Face at 85/100 AA. Anthropic followed at 62/100 BBB. NVIDIA held at 53/100 BB. DeepMind closes the month at 60/100 BBB, ranked last of 20 in the inaugural index. The transparency paradox has no cleaner illustration: the more valuable the talent, the less the market demands legible compensation information.
The scorecard
The ENTRA Pay Transparency Index scores five weighted dimensions. DeepMind's 60 breaks down as follows (ENTRA Pay Transparency Audit Q2 2026, data window January to June 2026; posting audit July 2026):
Salary Band Disclosure: 18 of 30. The index's most heavily weighted dimension. UK postings show salary ranges with reasonable consistency, ERA compliance pressure pulling disclosure rates above the global DeepMind baseline. California and Colorado roles carry bands per state law. Global research roles, including positions posted without a fixed geography and non-regulated US cities, routinely omit any range. A researcher applying to a Mountain View role may see a band. A researcher applying to an equivalent posting tagged only to a research org typically does not.
Equity Compensation Transparency: 12 of 25. DeepMind, as a wholly owned Google subsidiary, grants equity in Alphabet stock (GOOGL). Cliff, vesting schedule, and grant size by level are disclosed at offer stage. They are not public before it. Candidates cannot benchmark the equity component before the first recruiter call. That single gap, the inability to model equity value before entering the process, drives this score below both Hugging Face (21/25) and Anthropic (estimated 15/25).
Internal Pay Equity Data Published: 9 of 20. DeepMind UK files a Gender Pay Gap report under the mandatory UK framework, which applies to organizations with 250 or more employees. The 2024/25 filing shows a mean gap of approximately 21 percent, three points above the UK technology sector median of 18 percent. An AI-specific demographic breakdown, which would isolate the research org's pay distribution from the broader DeepMind population, is absent. The legal obligation is met. The voluntary specificity that earns additional index credit is not.
Compensation Philosophy Public: 14 of 15. The strongest single-dimension result. DeepMind and Google publish "world-class compensation" language in talent-facing materials. That phrasing earns near-maximum credit. The named percentile target, a statement of the form "we pay at the Xth percentile of the global technology market," which would close the final point, does not appear in public-facing documentation.
Market Commitment and External Benchmarking: 7 of 10. Levels.fyi aggregates data for DeepMind roles (Q2 2026), providing candidates a reference point that exists in practice if not by official endorsement. DeepMind has not officially linked to or validated this benchmarking. The score reflects acknowledged but not validated external data.
Hugging Face, at 85/100 AA, posts salary ranges globally without legal compulsion and across every geography where it hires. Anthropic, at 62/100 BBB, discloses US base bands while leaving equity mechanics undisclosed pre-offer. DeepMind, at 60, sits two points below Anthropic, with equity transparency as the single largest downward pull on the composite score.
The research bargain
The candidates who receive DeepMind offers do not, in most cases, open negotiations by asking for the salary band. They ask about compute allocation.
Access to Gemini training infrastructure is not a benefit listed in a recruiter email. It is the primary attractor for a specific category of researcher whose work requires scale that no startup can provide and that only a handful of organizations globally can deliver. AlphaFold 3 required infrastructure decisions at a scale that produced a Nature publication. Project Astra runs on production systems that DeepMind researchers can instrument directly. The h-index opportunity inside the DeepMind research org is grounded in that access, not in the posted band.
The research bargain, stated plainly: a candidate who can generate a DeepMind offer accepts that the compensation conversation happens inside the offer process, not before it. They accept that equity mechanics will be disclosed at offer acceptance, not at application. They accept opacity in exchange for compute access, research scope, and a publication environment that no transparency-maximizing startup can replicate. ENTRA's Q2 2026 recruiter network data (n=31 DeepMind London placements, anonymized, H1 2026) found zero candidates who cited pay transparency as a top-five factor in accepting or declining a DeepMind role. Every candidate cited compute access, research scope, or publication environment as the deciding variable. The research bargain is not a cultural accident. It is a market equilibrium, stable because the candidates who can clear DeepMind's bar are also the candidates least dependent on pre-application information to evaluate an offer.
What the data reveals
DeepMind UK's 2024/25 Gender Pay Gap filing is legally required and publicly available. The mean gap of approximately 21 percent sits above the UK technology sector median. The driver is seniority concentration: senior research roles, which carry higher compensation, are disproportionately held by men. This is not a finding unique to DeepMind, but the GPG report carries no AI-research-specific analysis that would allow a candidate or regulator to separate the research org's pay distribution from the broader organization.
The salary visibility pattern tracks jurisdiction. London-based Research Scientist postings in the Q2 2026 audit window showed £130,000 to £220,000 in base salary; Senior ML Engineers ran £110,000 to £175,000. Mountain View Senior Research Scientists ran $175,000 to $280,000 per Levels.fyi Q2 2026. These figures exist in the public domain. They are assembled from ERA-driven compliance postings, third-party aggregation on Levels.fyi, and ENTRA's posting audit, not from a DeepMind voluntary disclosure architecture.
The ERA 2025 countdown
The Employment Rights Act 2025 requires UK employers with 250 or more employees to publish pay equity action plans from April 2027. DeepMind UK, as part of a global organization of 3,500-plus employees with a London research workforce that has grown through Project Astra and Gemini infrastructure expansion, qualifies. Pay equity action plans under the ERA 2025 framework require narrative disclosure: what the gap is, what drives it, and what the organization commits to doing about it. The GPG filing addressed the what. The ERA 2025 action plan will require the why and the plan, in a research org where the mean gap runs three points above sector median.
This is the first regulatory forcing function that directly targets DeepMind's transparency gap. The voluntary disclosure decisions that have held the score at 60 may shift when a legal narrative requirement arrives in April 2027.
Path to AA
The distance from 60 to AA tier (80 and above) is 20 points. ENTRA's editorial projection: extending salary band disclosure globally, posting ranges on all roles rather than only where California, Colorado, or ERA compliance compels it, could add 8 to 12 points on the index's most heavily weighted dimension. Publishing an AI-specific demographic breakdown alongside the mandatory UK GPG filing could add 5 points on Internal Pay Equity. Combined: a realistic composite of 73 to 77, approaching A tier, with AA achievable by Q1 2027 if ERA 2025 preparation drives broader voluntary disclosure posture.
Whether the research bargain culture permits that shift is the open question. A research organization where the negotiating variable is compute allocation, not salary band range, has weak internal incentive to increase pre-application transparency. The April 2027 deadline may produce the first genuine forcing function the market has not.
Salary Transparency Month August 2026 closes with DeepMind as its starkest illustration. The organization whose research sets the frontier standard for AI ranks last on willingness to make compensation legible before the offer stage. DeepMind does not need transparency to attract talent. That is exactly what the index was built to surface.
ENTRA Pay Transparency Index August 2026 methodology: five weighted dimensions, Salary Band Disclosure (30 pts), Equity Compensation Transparency (25 pts), Internal Pay Equity Data Published (20 pts), Compensation Philosophy Public (15 pts), Market-Rate Commitment (10 pts); 20 AI employers scored; data window January to June 2026; posting audit July 2026; BBB tier: 60 to 69; AA tier: 80 to 89. Google DeepMind score breakdown: Salary Band Disclosure 18/30, Equity Compensation Transparency 12/25, Internal Pay Equity Data Published 9/20, Compensation Philosophy Public 14/15, Market-Rate Commitment 7/10; total 60/100 BBB. Salary data: Mountain View Senior Research Scientist $175K to $280K base per Levels.fyi Q2 2026; London Research Scientist £130K to £220K base and Senior ML Engineer £110K to £175K base per ENTRA Q2 2026 posting audit, July 2026 window. UK Gender Pay Gap: DeepMind UK 2024/25 filing; mean gap approximately 21%; mandatory UK filing requirement for 250-plus employee organizations; UK technology sector median 18% per ENTRA benchmark. Google DeepMind global headcount: approximately 3,500-plus per public sources. Employment Rights Act 2025: pay equity action plans required April 2027 for UK employers with 250-plus employees. ENTRA Q2 2026 recruiter network data: n=31 DeepMind London placements, anonymized, H1 2026. Comparison scores: Hugging Face 85/100 AA (ENTRA Pay Transparency Index August 2026, briefing published 2026-08-16); Anthropic 62/100 BBB (ENTRA Pay Transparency Index August 2026, briefing published 2026-08-09); NVIDIA 53/100 BB (ENTRA Pay Transparency Index August 2026, briefing published 2026-08-23). Individual named companies did not confirm ENTRA Pay Transparency Index scores.
Global Career PlatformFind AI talent. Find your next role.
Booking is hotels. · Airbnb is apartments. · ENTRA is global careers.