There is no new-grad research engineer program at Google DeepMind, so a master's student gets in through one of two side doors: the Student Researcher program while still enrolled, or one of the few research engineer postings whose bar a fresh graduate can clear. On 17 October 2026, DeepMind had 111 open postings on Google's careers site, 17 of them matched "research engineer", and only four were realistic for someone finishing a master's.
The mistake most applicants make is treating DeepMind like a Google new-grad pipeline. Filtering DeepMind's postings to Google's "Early" career level returned zero results that day. You are applying to specific teams with specific requirements, and the requirements decide everything.
Who they actually hire
Here are the 17 research engineer postings from 17 October 2026, grouped by what their minimum qualifications ask for:
| Minimum bar in the posting | Postings | Examples |
|---|---|---|
| No degree, no prior work experience | 1 | Polaris (Mountain View): build eval tasks that test Gemini on real software engineering |
| Master's degree as the floor | 2 | Agentic Security for Gemini (MS + 1 year training or fine-tuning generative models); Responsible Frontier AI Research (London, MS or PhD + AI safety research experience) |
| Bachelor's + 2 years of ML | 1 | Biosecurity (London) |
| Bachelor's + 3 to 8 years, or a PhD + 5 years | 13 | Energy Grids, Cyber Gemini, Frontier Safety Mitigations, staff-level post-training roles |
Source: Google careers, DeepMind filter, query "research engineer", retrieved 17 October 2026.
Read the four open ones literally. The Agentic Security role wants one year of hands-on training or fine-tuning of generative models plus experience in ML safety, security, privacy or alignment. That year can be a master's thesis or a research assistant role if you actually trained models. Responsible Frontier AI Research asks for prior AI safety or responsible AI work and lists Python and C++ for high-performance ML libraries as preferred. Polaris is the outlier: the posting says there are no education or degree requirements and no prior work experience is required, prefers olympiad or Putnam-level results, and says a single well-designed eval task takes an engineer about a week.
The pattern is clear. DeepMind hires early-career research engineers for evidence, not for credentials: a model you trained, a safety evaluation you built, a competition result.
The Student Researcher route
If you are still enrolled, this is the main door. DeepMind's program page describes paid, in-person placements of 12 to 24 weeks for bachelor's, master's and PhD students, with applications grouped by degree and location. On 17 October 2026, Google's careers site listed BS/MS postings for Fall 2026 in the US and Winter-Summer 2027 in Montreal, plus a 2027 posting in Zürich open to all degree levels.
Three details from the postings matter for master's students:
- PhD-first, in writing. The Winter-Summer 2027 posting says most projects are designed for PhD talent, and that bachelor's and master's places are "highly limited" and need a demonstrated research focus or a strong publication record.
- One application, many teams. You are considered for projects across Google DeepMind, Google Research and Google Cloud. A Google Research placement is not a consolation prize; it is the same pool.
- Paperwork and patience. Applications need a CV with your anticipated graduation date and a transcript in English. Review is rolling and "may take several months", and start dates for the 2027 cycle typically run from January to August 2027.
The US Fall 2026 BS/MS posting lists a full-time-equivalent base range of $98,000 to $131,000 and 19 US office locations. You must be physically in the posting's country for the whole placement, so check where your visa lets you work before you apply.
The process, step by step
DeepMind does not publish its research engineer loop. This table combines techinterview.org's July 2026 guide with candidate accounts on Blind.
| Stage | Format | What is graded | Timing |
|---|---|---|---|
| Recruiter screen | Call | Track (RE vs RS vs SWE), level, background | Week 1 |
| Hiring manager screen | Call | Fit with the specific team's work | Week 1-2 |
| Technical screen | 1-2 rounds | Coding, or an ML probe | Week 2-3 |
| Core loop | 5-7 rounds (guide); 4 core rounds in a 2024 Blind account | ML coding, standard coding, ML system design, ML fundamentals, behavioral | Week 3-6 |
| Hiring committee and team match | Written packet review | Consistency across rounds | 6-10 weeks total |
techinterview.org says AI tools are generally prohibited or heavily limited in technical rounds and that the RE loop includes implementing ML components unaided, distributed-training design and evaluation-infrastructure design. Ask your recruiter which version you are getting.
Candidate accounts fill in the texture. An August 2024 Blind poster listed two coding rounds, one ML system design and one ML fundamentals. A reply said the ML fundamentals round felt like DeepMind's old quiz: classic topics such as SVMs, naive Bayes and decision trees next to convolutions, transformers and optimisers, with light maths. One coding round was a medium-to-hard algorithms problem; the other was software fundamentals such as class design, multithreading and memory management. The same reply warned that the hiring-manager conversation can turn technical.
The famous quiz has changed, not disappeared. In a June 2023 Blind thread, candidates first argued about whether the maths and statistics questions were gone, then agreed they still come up, just not as the earlier two-hour exam-style round. An older November 2022 account of an L3/L4 loop described a hard dynamic-programming question mixed with probability, and verbal questions on Python's GIL and low-level NumPy optimisation.
What gets people rejected
1. Applying to senior postings and hoping. Thirteen of the 17 research engineer postings ask for three or more years, or a PhD plus five. A master's with no full-time experience will not get past the screen for those.
2. No trained model on the CV. A 2021 Blind poster said they were rejected for not having enough ML experimentation and modelling experience. Coursework does not count as the year of fine-tuning the Agentic Security posting asks for.
3. Preparing only LeetCode. The 2024 account had a whole round on ML fundamentals and another on software fundamentals. Being unable to explain why Adam differs from SGD fails a round your algorithms prep never touched.
4. A thin research story. On a levels.fyi thread, a new grad with a CS master's who landed an RE interview was told to prepare a story for every resume bullet and to lead with the impact of their research, because the people interviewing them often hold PhDs.
5. Treating a Student Researcher application as quick. Review can take months. Apply early and keep other processes alive.
The visa question
In the UK, the Home Office register of licensed sponsors dated 16 October 2026 lists Google (UK) Limited in London with an A (Premium) rating for Skilled Worker sponsorship, and no separate DeepMind entity. None of the postings we read stated a sponsorship policy either way, so ask the recruiter directly.
The Graduate visa clock also shortens soon. GOV.UK says the visa lasts two years if you apply on or before 31 December 2026 and 18 months if you apply on or after 1 January 2027; PhD graduates get three years. A master's student finishing in 2027 gets the shorter window, which leaves less time for a several-month DeepMind process. Our UK quant sponsorship guide covers how to read the register.
A 4-week plan
- Week 1: Pick your door. Still enrolled: apply to the Student Researcher posting for your country now. Graduating: shortlist the four realistic research engineer postings above and read their minimum qualifications line by line. Run your CV through the free ATS resume checker and tighten it with the Google resume polish page, since DeepMind applications run through Google's careers system.
- Week 2: ML fundamentals. One topic a day, explained out loud in five minutes: bias-variance, regularisation, SVMs, attention, optimisers, evaluation leakage.
- Week 3: Implement unaided. Write a transformer block, a training loop and a data pipeline from scratch in PyTorch or JAX, with no AI assistant. Then do one medium-to-hard algorithms problem a day and one class-design problem with threading.
- Week 4: Mock the loop. Applr's DeepMind research engineer mock interview runs ML-fundamentals and coding follow-ups in this format. For the general Google coding bar, add the Google L3 software engineer mock.
Who should aim where
- Master's student still enrolled, with a paper or lab work: Student Researcher, then convert.
- Graduating master's who fine-tuned models in a thesis: Agentic Security for Gemini or Responsible Frontier AI Research.
- Strong competitive programmer, any degree: Polaris.
- No research output yet: compare doors with our Anthropic new grad guide and OpenAI new grad guide, and track openings on the New Grad 2027 tracker.
Sources
- Google Careers, DeepMind postings matching "research engineer", retrieved 17 October 2026
- Google Careers, "Research Engineer, Agentic Security for Gemini, DeepMind", retrieved 17 October 2026
- Google Careers, "Research Engineer, Polaris, DeepMind", retrieved 17 October 2026
- Google Careers, "Research Engineer, Responsible Frontier AI Research, DeepMind", retrieved 17 October 2026
- Google Careers, "Student Researcher, BS/MS, Winter-Summer 2027", retrieved 17 October 2026
- Google Careers, "Student Researcher, BS/MS, Fall 2026", retrieved 17 October 2026
- Google DeepMind, Student Researcher Program, retrieved October 2026
- techinterview.org, "Google DeepMind Interview Process 2026", updated 12 July 2026
- Blind, "Deepmind research engineer interview help", August 2024
- Blind, "Interviews for Research Engineer at DeepMind", June 2023
- Blind, "DeepMind research engineer loop", October-November 2022
- Blind, "DeepMind Research Engineer Interview process", April 2021
- Levels.fyi community, "DeepMind research engineer interview", retrieved October 2026
- Home Office, Register of licensed sponsors: workers, 16 October 2026
- GOV.UK, Graduate visa, retrieved October 2026
Frequently asked questions
- Can you get into Google DeepMind with a master's degree?
- Yes, but not through a new-grad program, because DeepMind does not run one. On 17 October 2026, 17 DeepMind postings on Google's careers site matched 'research engineer', and four of them were realistic for someone finishing a master's: two set a master's as the floor (Agentic Security for Gemini and Responsible Frontier AI Research), one had no degree or experience requirement (Polaris), and one asked for a bachelor's plus two years of ML (Biosecurity). The other route is the Student Researcher program while you are still enrolled.
- What is the DeepMind research engineer interview process?
- DeepMind does not publish it. techinterview.org's July 2026 guide describes a recruiter screen, a hiring-manager screen, one or two technical screens, a five-to-seven-round loop, then hiring committee and team match, over six to ten weeks. Candidate posts on Blind from 2024 describe four core rounds: two coding, one ML system design and one ML fundamentals.
- Does DeepMind still have the math and statistics quiz?
- Not in its old form, according to candidates. In a June 2023 Blind thread, people preparing for the research engineer loop agreed that maths and statistics questions still appear, but the earlier two-hour exam-style round has changed. A 2024 Blind reply described the ML fundamentals round as covering classic topics such as SVMs, naive Bayes and decision trees alongside transformers and optimisers, with light maths.
- How much does a DeepMind research engineer earn?
- US postings publish ranges. On 17 October 2026, Research Engineer, Agentic Security for Gemini listed a $174,000 to $252,000 base plus a 15% bonus target and equity, and Research Engineer, Polaris listed a $164,000 to $252,000 base with $550,000 to $1,100,000 total compensation including bonus, equity and sign-on. London postings did not show a range.
- Can master's students apply to the DeepMind Student Researcher program?
- Yes, but the 2027 postings warn that most projects are designed for PhD students and that bachelor's and master's places are highly limited and need a demonstrated research focus or strong publication record. One application is considered across Google DeepMind, Google Research and Google Cloud, and the BS/MS Winter-Summer 2027 posting asks for a transcript and an anticipated graduation date.