Job search
Research Engineer vs. Research Scientist Jobs: Which Requirements Should Shape Your Search?
Compare research engineer and research scientist jobs by research ownership, engineering work, publication evidence, and each posting's actual requirements.
Research Engineer and Research Scientist are useful search terms, but they are not reliable job descriptions by themselves. Choose the posting whose expected work and qualifications match your strongest evidence: what question you will help answer, what system or experiment you will own, and how the team expects you to show that you can do it. Research and engineering overlap in many labs, and some employers combine both titles. A PhD may be central to one opening and unnecessary for another.
LandOffer publishes this guide. You can browse recent roles on LandOffer, then verify every requirement on the employer's own posting. The examples below are live Anthropic role descriptions and Google DeepMind's current job-family guidance checked October 13, 2026. They illustrate specific employers, not an industry-wide rule or a prediction of who will be hired.

Start with the work you will own
Read the responsibilities before deciding what the title means. A research scientist posting may ask you to frame questions, choose methods, interpret ambiguous results, and communicate what the evidence supports. A research engineer posting may ask you to implement experiments, build systems that make them repeatable, scale compute, improve performance, and help researchers test ideas. Those are tendencies in some job families, not a universal division of labor. In practice, a scientist may write substantial code, while an engineer may shape a research direction or co-author a paper.
The Google DeepMind careers page describes its Research Engineers as software engineers with deep ML understanding who design, build, and scale systems to test ideas. Its Research Scientists formulate hypotheses, evaluate models, and often contribute to foundational papers; it says they normally hold a PhD. “Normally” describes the job family rather than a mandatory rule for every vacancy. Anthropic offers a useful counterexample to rigid title assumptions: its career guidance says engineers do research, researchers do engineering, and technical backgrounds vary.
A practical first pass is to finish this sentence for each vacancy: “The person hired will be trusted to ___, and the evidence I can show is ___.” If you cannot name the responsibility, revisit the posting. If the evidence is hard to connect to the output, do not fix the mismatch by repeating keywords; identify what is transferable and what remains unproven.
Compare two postings from the same team
Two Anthropic Interpretability vacancies show why a same-employer comparison is more useful than a dictionary definition. The Research Engineer, Interpretability posting centers on specialized training and inference infrastructure, profiling, performance, tooling, and translating research needs into engineering solutions. It asks for substantial software-building experience and says previous research experience is not required. The Research Scientist, Interpretability posting emphasizes scientific research, experimental design and analysis, interpretability work, and communication of results, including null findings.
The scientist application requests a publications or research-output link and tells candidates without publications to apply to the Research Engineer posting instead. This is a concrete, role-specific signal about the evidence that team is seeking. It is more informative than assuming that every scientist vacancy requires a doctorate. In the logistics section, both postings list a bachelor's degree or an equivalent combination of education, training, or experience as minimum education. The role-specific evidence is different: scientific record and research output on one side, software and infrastructure experience on the other.
| What to compare | Research Engineer example | Research Scientist example | What to verify in your posting |
|---|---|---|---|
| Main output | Reliable research infrastructure, tools, scaling, and performance work | Research methods, experiments, interpretation, and communicated findings | What must be designed, built, run, or explained? |
| Evidence of fit | Software systems, programming, profiling, infrastructure, or related work | Research record, methods, results, domain experience, or publications | Which evidence does the required section name? |
| Prior research | The example explicitly says prior research experience is not required | The example asks for scientific research track record and some interpretability work | Is research experience required, preferred, or simply relevant? |
| Publications | The example asks for engineering and technical achievements; no publication link requirement appears | Application requests a publications/research-output URL | Is a paper mandatory, preferred, or one way to demonstrate the work? |
| Degree signal | Standard minimum education says bachelor's or equivalent | Same standard minimum education, with different role-specific research expectations | Does this particular role say PhD required, preferred, normally held, or not at all? |
This comparison does not imply one job is more senior, more prestigious, or easier to enter. The posted engineer role asks for five to ten or more years of software-building experience, while the scientist role asks for a strong research record and prior interpretability work. Compare the whole qualification set and level; do not replace it with an imagined ladder in which engineers are always entry-level and scientists always hold doctorates.
Treat degree and publication language precisely
A posting may use several kinds of qualification language. A required minimum is a direct screen. A preferred qualification can strengthen an application but is not written as a prerequisite. “Normally holds a PhD” is a job-family description, not the same as a role-specific PhD requirement. “Equivalent experience” may broaden how a qualification can be met, but it does not erase other required evidence. An application field asking for a publication link may signal that the team expects research output; read any alternative or referral instruction exactly as written.
Make a small evidence map instead of guessing what a credential means:
| Posting language | Your check | Example response for your own planning |
|---|---|---|
| “PhD required” | Is it an explicit minimum, and does the listing allow equivalent experience? | “The posting lists a PhD as required and provides no equivalency. I need to clarify before treating experience as a substitute.” |
| “PhD preferred” or “normally holds a PhD” | Which specific research skill does the degree appear to represent? Can another work sample demonstrate it? | “The team values advanced research training; my peer-reviewed project and methods write-up may be relevant, but I will compare them with the actual must-haves.” |
| “Strong research track record” | What completed output shows the question, method, result, contribution, and limitations? | “I can link a paper and explain my portion, including a result that did not support the initial hypothesis.” |
| “Publication link” | Is a publication required, or does the form accept another research output? | “The form explicitly directs candidates without publications to another role. I should follow that instruction rather than submitting an unrelated link.” |
| “Research experience not required” | Which engineering evidence still must be present? | “I can prioritize systems work if I can show software depth and communicate why the team’s research matters.” |
A publication list is not interchangeable with proof that you performed the work described. If you are a co-author, state your actual contribution and be ready to explain the methods, decisions, and limitations. A dissertation, preprint, technical report, open-source artifact, experiment log, or production result may be relevant when a posting allows it and when you can clearly explain your role. Do not label an unpublished manuscript as a published paper or imply that a repository contribution was independent research if it was not.
Notice when a role deliberately combines both
Some postings are not a choice between scientist and engineer. Anthropic's Research Engineer / Research Scientist, RL Frontiers posting describes work on algorithms and model architectures, rigorous experiments, large-scale implementation, training systems, cost and performance models, and diagnosing failures at scale. Its listed qualifications include both ML training research and systems expertise. A candidate who searches only one title could miss a role whose responsibilities span both.
A combined title also raises a useful interview question: which outcomes will define success in this particular team? Ask whether the role is expected to originate algorithms, build the infrastructure that enables experiments, own both ends of a research cycle, or move among those responsibilities as the project changes. The posting may answer some of this, but a recruiter or hiring manager can clarify the balance, the first project's scope, and the evidence they want to see.

Match your strongest evidence to the role
Consider two fictional candidates reviewing the same set of postings. These are teaching examples, not real applicants or predictions.
Maya has built distributed data and model-training systems, diagnosed throughput regressions, and maintained a reproducible evaluation pipeline. She has contributed to research teams but has no paper to link. The Interpretability Research Engineer opening is a plausible first application because its description values software infrastructure and says prior research experience is not required. She should lead with a system she owned, what constraints it faced, how she measured performance, and how researchers used it. The Research Scientist opening is a weaker fit for this particular application because it requests a research track record and output link; she should not invent one or assume her engineering work meets that requirement. If her experience also spans algorithm development and experiment design at scale, the combined RL posting may merit a separate review.
Eli has led experiments in computational neuroscience, published a paper, and can explain the study design and limitations. He writes analysis code, but has not built or operated large software systems. The scientist opening may better match his documented research outputs if he can show relevant interpretability work. He should describe which parts of the paper he designed and executed rather than relying on its title or author order. The engineer opening still requires serious software experience; “I can code for my research” alone may not establish the posting's stated software-building depth. A research-focused role in another scientific domain could fit better if interpretability work is only a curiosity rather than a demonstrated area.
The choices change because the evidence changes, not because one candidate holds a preferred identity. If your record straddles both, apply to both only when each application can make a distinct, truthful case against that posting's requirements. One generic resume that lists papers and tools without showing your contribution may undersell both profiles.
Screen a research job in fifteen minutes
Use this sequence before you invest in a tailored application:
- Copy the exact title and level. Search for neighboring titles too: Research Engineer, Applied Scientist, Research Scientist, ML Engineer, Research Software Engineer, and combined variants. The title is a search clue, not a qualification.
- Circle hard requirements. Capture degree, experience range, domain, methods, location, work authorization, and any application item that blocks submission. Separate explicit minimums from preferred qualifications.
- Underline the verbs in the responsibilities. “Formulate,” “design,” “measure,” “interpret,” “build,” “scale,” “deploy,” and “maintain” reveal what the employer expects the person to own. Count the work products, not just the nouns.
- Choose two or three requirements to prove. For each, name a completed project, paper, system, experiment, or tool and your actual contribution. If you cannot point to evidence, mark the gap; do not turn interest into experience.
- Check the application itself. A required publication URL, work sample, essay, or portfolio field can reveal what the team will evaluate. Follow explicit alternatives and submission instructions; do not upload unrelated or confidential work.
- Resolve practical constraints. Confirm current location, office expectations, travel, authorization and timing on the employer page. These details can change, and an employer's general careers guidance may not apply to every role.
- Make a decision. Apply when you can substantiate the core requirements and explain your contribution. Ask one focused question when an essential term is unclear. Move on when a stated hard requirement cannot be met and the posting offers no alternative.
This screen is a prioritization tool, not an eligibility ruling. Employers can change a vacancy or evaluate equivalent evidence differently. When a posting explicitly invites applicants who miss some preferred qualifications, use that instruction; when it states an unambiguous hard minimum, do not rewrite it as optional.
Build a shortlist around work, not labels
A useful shortlist can include both role families. Save each opening with the expected output, named requirements, evidence you will use, unresolved questions, location, and application destination. That makes an overlapping title easier to evaluate and helps you avoid applying to an unrelated role just because it contains “research.”
For example, a candidate with a new paper and strong experimental design but little software ownership may prioritize research-scientist or applied-science postings. An engineer with proven large-scale systems, tooling, and close research collaboration may prioritize research-engineering roles, including openings that say prior research is unnecessary. Someone with both algorithm and infrastructure work can compare combined roles directly. Revisit this allocation as your work samples change; the categories describe vacancies, not permanent limits on your career.
If the posting names a PhD, publication, or years-of-experience requirement, copy its exact wording into your shortlist and record whether it is required, preferred, typical, or a form prompt. Then choose the closest recent work you can explain from question through result. That is a more useful basis for the next application than assuming a title determines your eligibility. To build that shortlist, browse recent roles on LandOffer, then open the employer posting for each role you plan to pursue.


