AI research scientist vs research engineer: what's the difference?

In the research scientist vs research engineer split, the scientist decides what to try: the question, the experiment, the paper. The engineer makes it work at scale: the training code, the data pipelines, the GPUs. At today's AI labs the line is blurry, titles vary by company, and the PhD matters less than its reputation suggests.

To check, we read the full text of every AI research ad on Tokn as of 14 October 2026: 95 roles at 26 companies in Europe, selected by job title. Research is one of the main AI career paths, and the one with the most confusing titles.

In short

• Scientists own the research question; engineers own the systems that test it.
• On Tokn there are about two research engineer ads for every research scientist ad (42 vs 23).
• Only 3 of 85 permanent research ads strictly require a PhD. 51 never mention one.
• CUDA, kernels, C++ and Rust mark engineer ads; a PhD and named conferences mark scientist ads.
• The UK (33 ads) and France (31) lead, before counting Google DeepMind and others we don't track.

Research scientist vs research engineer: the short answer

A research scientist designs the experiments; a research engineer builds the systems that run them. Google DeepMind keeps the classic split: on its careers page, research scientists identify "core research questions" and "normally hold a PhD", while research engineers are "software engineers with a deep understanding of machine learning" who build and scale systems and "often co-author research papers".

The cleanest pair we found was at Datadog's AI research team in Paris, advertised in October. Scientists work "on fundamental research problems"; engineers "turn research ideas into working systems, building the data, tooling, and infrastructure that enable rapid iteration". For the interview side, see our guide to machine learning interview questions.

What each one does day to day

Scientists spend more time on ideas and analysis, engineers on code and infrastructure, but both write code. An OpenAI "Researcher, Training" ad in London put it plainly: the right person is "equally happy coming up with a creative breakthrough, investing in strengthening a baseline, designing an eval, debugging a thorny regression, or tracking down a bottleneck." A Wayve applied scientist ad listed a typical day: coding and debugging models, pairing, reviewing pull requests and experiment designs, planning experiments, and reading groups.

Research scientist

A scientist frames the problem and owns the result. A Synthesia ad asked scientists to "own well-scoped research problems end to end"; Datadog's wanted "a track record of impactful publications at top-tier venues".

Research engineer

An engineer either builds the shared platform or sits inside a research team. Mistral AI's ad described both: a platform team that improves "the shared training framework, data pipelines and cluster tooling used by every team", or an embedded team that sits inside a research squad and turns "fresh ideas into repeatable, scalable code".

The research areas hiring in Europe

Counted from job titles on Tokn: post-training and reinforcement learning (13 ads), inference and research infrastructure (about 13), multimodal, image and video (about 11), pre-training (6), safety and alignment (4), science and health (4), evaluation (3). Interpretability had no titled role.

Applied scientist, Member of Technical Staff, RE/RS: decoding lab titles

Each lab draws the line differently, so read the description, not the title. Here is how European ads used them in October.

"Member of Technical Staff" has no official definition from any lab. Fast Company and Simplify describe a flat title that says nothing about seniority.

Do you need a PhD for AI research?

Usually not. Of 85 permanent research ads on Tokn, only 3 required a PhD with no alternative, and 51 (60%) never mentioned one. We coded each ad by hand.

The three strict ones were at Shift Technology and Spotify. The gap between tracks is real: 58% of scientist-side ads mention a PhD, against 30% of engineer-side ads.

Some labs say it outright. An Anthropic reinforcement learning research engineer ad in London read: "Strong candidates need not have: Formal certifications or education credentials; Academic research experience or publication history". ElevenLabs' research engineer ads: "We do not require any formal certifications or degrees." The counterpoint is DeepMind, whose research scientists "normally hold a PhD".

We found no reliable European data on PhDs in AI research hiring. In North America, the Stanford AI Index 2026 reports that new AI PhDs in the US and Canada rose 22% from 2022 to 2024, but that the increase went to academia, not industry.

Do publications matter?

Less than you might fear: 26% of ads name a venue such as NeurIPS, usually with alternatives. Aleph Alpha asked for "top-tier publications (NeurIPS, ICML, ICLR), impactful open-source contributions, or significant shipped technical work". Some engineer ads push the other way: Synthesia wanted people who "care about shipping, not just publishing."

Skills that separate the two tracks

PyTorch is the common ground; systems skills separate engineers. These shares come from a keyword search of full ad text, so they are approximate.

JAX appeared in about 1 in 4 ads, usually as "PyTorch or JAX", so PyTorch is the safe first choice. Engineer ads name FSDP, DeepSpeed, Megatron-LM and SLURM. For the wider market, see the most in-demand AI skills.

How much do AI research scientists and engineers earn in Europe?

There is no solid European data. Tokn read a salary on only 11 of 85 permanent research ads, all at three frontier labs: Anthropic, OpenAI and Black Forest Labs. Treat them as outliers, not typical pay.

We left out US and Canadian dollar bands shown on multi-city ads. Anthropic's two titles share a £260k floor, so nothing here shows one track paying more. For normal ranges, see typical AI salaries in Europe and machine learning engineer salaries.

How to get into AI research without a PhD

The research engineer role is the widest door, with 42 of the 95 ads. Mistral AI's ad said "Engineers can move along the research↔production spectrum as needs or interests evolve", and co-authoring papers is the usual way across. If you are not there yet, start by becoming an ML engineer.

Fellowships, residencies and internships

The Anthropic Fellows catch, per Anthropic's careers board, is eligibility: EU-only residents can't apply. See also DeepMind's student researcher page and OpenAI's residency page. PhD internships at Cohere, H Company and Helsing required PhD enrolment; Together AI's took Bachelor's and Master's students too.

Public work counts too. ElevenLabs asked for "past projects, designs, or GitHub contributions" instead of degrees, and 28% of permanent ads mentioned open source.

If you want a PhD anyway

The ELLIS PhD Program pairs advisors in two European countries and has an industry track. Its call for 2027 starts closes on 31 October 2026; 5–10% of eligible applicants are accepted in a typical round.

Where the AI research jobs are in Europe

The UK (33 ads) and France (31) dominate, then Germany (13), Spain (6, all Helsing in Barcelona) and Switzerland (4). Paris had 29 and London 23.

Mistral AI posted the most (17), then Anthropic (10), Helsing (9) and Synthesia (8). About 4 in 10 came from product companies such as Helsing, Synthesia and Datadog, not model labs. 19 were fully remote, mostly at Synthesia, Spotify, ElevenLabs and Cohere.

Tokn doesn't track Google DeepMind, Isomorphic Labs or Meta FAIR: check their careers pages directly. DeepMind alone listed at least 9 research scientist and about 5 research engineer roles in London, a scientist-heavy split unlike our board. We also miss some Cohere and Black Forest Labs research postings, so our counts for them are not their totals.

Which one should you choose?

Choose the scientist track if you want to set the research agenda and have, or want, a PhD. Choose the engineer track if you like making systems fast and reliable and want more openings (on our board, at least). You can switch later.

Find AI research jobs on Tokn

Browse AI research jobs in Europe, or research roles in Paris and in London. Research ads are few, so set up an alert for research roles.

FAQ

Do research engineers need a PhD?

No. Only 30% of engineer-side research ads on Tokn mention a PhD, usually with alternatives. Anthropic and ElevenLabs said explicitly that degrees are not required.

Who earns more, research scientists or research engineers?

No European evidence shows one paying more. At Anthropic, the only same-lab comparison on Tokn, both titles started at £260k, and frontier-lab ranges are outliers anyway.

Can a research engineer become a research scientist?

Yes. Mistral AI's ads say engineers can move along the research-to-production spectrum. The usual route is co-authoring papers, then taking on research direction.

Should I learn JAX or PyTorch for research jobs?

PyTorch. It appeared in 54% of permanent research ads, against 26% for JAX, usually as "PyTorch or JAX".

What does "Member of Technical Staff" mean?

A flat title that Cohere, Black Forest Labs and Perplexity use for both research and engineering roles. It says nothing about seniority, so read the description to see which way the job leans.