In [35]: # vs_research_scientist.ipynb
Abstract
At frontier AI labs, research scientists earn approximately 15 to 25 percent more than ML engineers at the L5 senior level. The gap is concentrated in equity grants, reflecting the lab's value placement on research-track contribution to foundation-model capability. The non-compensation differences are larger: research scientists require PhD credentials at almost all major labs, have higher publication freedom and external visibility expectations, lower direct product impact, and a narrower external market for role mobility. ML engineers have lower compensation at the same level but broader industry mobility and direct product-engineering scope [1].
1 Bands from Levels.fyi OpenAI, Anthropic, and arXiv author affiliations on recent frontier-AI papers, May 2026.
table rs-1 : MLE vs RS at frontier lab
| Dimension | ML engineer | Research scientist | Note |
|---|---|---|---|
| Base salary (L5 equivalent at frontier lab) | $280k - $400k | $300k - $450k | RS slightly higher |
| Equity grant (L5 equivalent at frontier lab) | $300k - $700k / yr | $400k - $850k / yr | RS noticeably higher |
| Total compensation L5 | $580k - $1.1M | $700k - $1.3M+ | RS approximately 15-25% higher |
| PhD requirement at intake | Common, not strictly required | Almost universally required | Different gates |
| Publication freedom | Limited at frontier labs (NDA, competitive) | Higher, with constraints | RS clearly more |
| Conference and external presence | Optional | Often expected as part of role | RS more |
| Direct product impact | Higher (productisation track) | Lower (research-track) | MLE more |
| Time-to-promotion at frontier lab | Standard L-track timing | Slot count smaller; competitive | MLE often faster |
| External market liquidity | Broad, many destinations | Narrower, frontier labs and academia | MLE more flexibility |
section rs-2 : non-compensation factors
The ML engineer vs research scientist decision is rarely compensation-driven, because the compensation gap at frontier labs (15 to 25 percent) is small relative to the gap between frontier-lab tier and other employer tiers. Both tracks at a frontier lab pay materially more than equivalent roles at hyperscalers or AI unicorns. The within-frontier-lab compensation delta usually does not drive the decision; the work content does.
The decision usually rests on four non-financial factors. First, work content: research scientist roles emphasise novel methodology and published findings; ML engineer roles emphasise shipped systems and direct product impact. Second, credential and background: PhD credential and demonstrated published research record open research-scientist roles; strong production-engineering background and demonstrated shipped ML systems open ML engineer roles. Third, recognition style: research-track recognition is external (academic publications, conference talks, named-author status); engineering-track recognition is internal (company-internal promotion, technical-leadership reputation, product-impact metrics). Fourth, long-term career optionality: research-track keeps academia as a viable future option; engineering-track keeps broader industry mobility as a viable future option.
A common pattern: senior ML engineers who entered industry without a PhD sometimes target a transition to research-scientist work mid-career, typically by accepting a small compensation cut to take a research-track role at a smaller frontier lab where the credentialing requirements are looser. The reverse pattern (research scientists transitioning to ML engineering) is rarer but happens, often when a research scientist finds that the publication-track recognition model is less personally rewarding than direct product-engineering impact.
section rs-3 : common questions
Does a research scientist make more than an ML engineer at a frontier lab?
Typically yes, by approximately 15 to 25 percent at the L5 senior level. Research scientist L5-equivalent at a top frontier lab earns total compensation $700,000 to $1,300,000+, while ML engineer L5-equivalent earns $580,000 to $1,100,000. The gap is concentrated in equity grants, which reflect the lab's value placement on research-track contribution to the lab's foundation-model capability. At hyperscaler ML organisations, the research-vs-engineering compensation gap is smaller, often only 5 to 10 percent at L5.
What is the difference between ML engineer and research scientist?
ML engineer (productisation track) focuses on shipping ML systems to users: pipeline-to-production ownership, feature store engineering, model serving infrastructure, experimentation framework design, and operational reliability for ML systems at scale. Research scientist (research track) focuses on novel methodology: foundation-model architecture research, scaling-laws empirical work, publication-track research contributions, and benchmark-pushing work. The two tracks overlap significantly in foundational skills (deep learning fundamentals, distributed training systems) but diverge in day-to-day work and in deliverable types (shipped product features vs published findings).
Do I need a PhD to be a research scientist?
Almost universally yes at frontier AI labs and at hyperscaler research organisations. Research scientist roles at the major labs are essentially closed to non-PhD candidates, with rare exceptions for candidates with extraordinary published research records or with senior demonstrated frontier-AI experience. The PhD requirement reflects the role's emphasis on novel methodology contribution: research-track work requires deep familiarity with the published literature and with the experimental methodology that PhD training develops. ML engineer roles (productisation track) are open to non-PhD candidates and often welcome strong industry-trained engineers.
Which track has better long-term career prospects?
Different rather than uniformly better. Research scientist track provides stronger paths into academic positions, into senior research leadership roles (head of research, lab director), and into publication-track external recognition. ML engineer track provides stronger paths into engineering leadership (VP of Engineering, CTO), into broader industry mobility (more employer options at senior level), and into product-engineering roles outside pure research. Many senior ML engineers cross between tracks during their career; the choice often reflects what kinds of problems and what kinds of recognition the engineer wants to pursue at senior career stage.
How does the publication freedom difference affect career trajectory?
Research scientists at frontier labs typically have higher publication freedom (papers can be submitted to top conferences with internal review), conference attendance and presentation expectations, and external visibility as part of the role definition. ML engineers at the same labs typically have lower publication expectations and lower external visibility, though some engineers do publish foundational systems work. The publication freedom difference compounds over a career: research scientists build external academic reputation that translates to optionality (academic positions, senior advisory roles, named-author status); ML engineers build internal company reputation that translates to compensation progression and to industry-internal mobility.
Can a research scientist become an ML engineer or vice versa?
Yes in both directions, with friction. Research scientists transitioning to ML engineering typically need to build production-systems engineering depth that academic research does not develop: distributed systems, infrastructure as code, operational reliability, real-world data pipeline messiness. The transition is most successful for research scientists who joined applied research labs (rather than pure-research labs) where some product-adjacent work was part of the role. ML engineers transitioning to research scientist roles typically need to build published research output: this requires 1 to 3 years of producing publishable work, which is harder to do without research-track time allocation and mentorship.
What is the salary at academic positions for former research scientists?
Academic tenure-track positions in CS or ML at top US universities pay materially less than frontier-lab research scientist roles: assistant professor starting compensation typically $130,000 to $200,000 in base salary plus 1 to 2 month summer salary, with limited equity and conventional retirement-plan benefits. Most former frontier-lab research scientists who move to academia accept a substantial compensation cut for non-financial reasons (research agenda independence, student mentorship, public-good work, tenure security). The reverse path (academic to industry) is common and well-compensated: a tenured CS professor moving to a frontier-lab research scientist role typically sees a 3 to 5x compensation increase.