EmergingReasoning, planning and search

Inference-Time Compute Researcher

Also called Test-Time Scaling Researcher

An Inference-Time Compute Researcher decides how much thinking a model should spend on each problem. They raise accuracy on hard tasks while keeping cost under control, which matters wherever answers must be right and budgets are finite.

What they do

The work starts with a model that sometimes fails on hard problems and wastes effort on easy ones. The researcher builds methods that spend more inference effort where it pays off. That might mean sampling more candidates, searching over reasoning steps, calling a verifier or running a symbolic solver. They design stopping policies that decide when to continue and when to give up. They run budget studies across task difficulty and compare methods at equal cost. They build search environments, check results against ground truth and write up what held and what did not. Much of the time goes on careful evaluation, not new ideas.

What a strong hire delivers

In the first months, a strong hire gives the team a clear picture of where extra inference effort helps and where it does not. They ship an adaptive inference method that beats a fixed budget baseline on verified success per unit of cost. They leave behind a budget study the team can rerun, a reusable evaluation setup and a plain recommendation on how to allocate compute across task types.

How you can tell they are good

Ask them to walk through a budget study they ran. Good candidates explain the baseline, the cost measure and how they checked that answers were truly correct. They can say when more search made results worse and why. Ask how they would decide to stop early. Weak candidates talk only about accuracy gains and ignore cost, or cannot separate a method's effect from a lucky sample.

Why it is hard to fill

The title barely exists yet. The work usually sits inside reasoning, post-training or applied research roles, so job titles tell you little. Candidates tend to be strong in one area, such as reinforcement learning or operations research, and thin in the others. Few have run cost-aware studies at scale. Searching by title misses most of them, so we look at the work people have actually published or shipped.

Where to find them

Look at researchers working on reasoning, planning and search inside model providers and research labs. Others come from operations research, combinatorial optimisation and algorithms groups, or from reinforcement learning teams that worked on game search. Some sit in applied teams building agents or code and maths solvers, where verified answers and cost both matter. Titles are often research scientist or ML engineer.

Indicative pay

Not enough published evidence yet to give a fair range for this role, which is common for work this new. We benchmark it for you as part of a search.

Interview questions for an Inference-Time Compute Researcher

  1. Describe a time you spent extra inference effort on a task. How did you decide where the budget went?

    Strong answers name the task, the allocation rule and the cost measure, and show a comparison against a fixed budget. Worry if they cite only accuracy, or if the allocation was a hand-tuned guess they never tested.

  2. How would you design a stopping policy for a model that samples several reasoning attempts?

    Look for signals such as agreement between samples, verifier scores and estimated difficulty, plus a view on the cost of stopping too early or too late. Worry about fixed thresholds with no validation on held-out tasks.

4 more questions, with what a strong answer covers, are in the Role Kit.

Job description you can post

We are building systems that reason through hard problems, and we want to spend inference effort where it counts. You will research and build adaptive methods that raise verified success while keeping cost in check. The work is new, so you will help define how we do it.

What you will do

  • Design methods that allocate extra inference effort to difficult tasks
  • Build and test stopping policies for search and sampling
  • Run budget studies comparing methods at matched cost
  • Build search environments and evaluation setups with reliable ground truth
  • Work with verifiers and symbolic solvers to check answers
  • Write clear reports with recommendations for the wider team

What you will bring

  • Shipped or published work on search, reasoning or planning with language models
  • Experience running cost-aware evaluations with clear baselines
  • Strong Python and comfort with experiment tooling
  • Working knowledge of transformer models and reinforcement learning
  • Grounding in optimisation, algorithms or operations research
  • Habit of testing whether a result is real before reporting it

Nice to have

  • Experience building or auditing verifiers
  • Background in combinatorial search or symbolic solving
  • Experience with production latency and cost constraints

Free to use and adapt. If it saves you time, a link back to this page is appreciated.

Skills that matter

  • Transformer modellingAdvancedImplement attention-based sequence or multimodal models
  • Reinforcement learningAdvancedOptimise policies from interaction and reward
  • Optimisation and operations researchWorkingChoose actions subject to objectives and constraints
  • Model evaluationWorkingMeasure capability and failure modes credibly

Other roles in reasoning, planning and search

EmergingReasoning Research ScientistA Reasoning Research Scientist tests and builds methods that make a model's multi-step inference reliable. They show the company which approaches…Hiring nowPlanning Algorithms ResearcherA Planning Algorithms Researcher builds planners that choose sequences of actions under hard constraints. Companies hire them so agents and systems…EmergingNeuro-Symbolic AI ResearcherA Neuro-Symbolic AI Researcher combines learned models with symbolic rules and solvers, so systems reason consistently where rules matter. That…

Related roles in other families

Hiring nowPost-Training Research EngineerPost-training and preference optimisationA Post-Training Research Engineer adapts pretrained models into ones that behave as the company needs. They turn preference data and reward signals…Hiring nowReward Modelling ResearcherPost-training and preference optimisationA reward modelling researcher builds models that learn what good output looks like from preference comparisons. They give post-training teams a…Hiring nowReinforcement Learning Research EngineerPost-training and preference optimisationA Reinforcement Learning Research Engineer improves how a model behaves by training it against interactive signals. They matter because reward gains…Hiring nowNatural Language Processing ScientistLanguage and multilingual intelligenceAn NLP Scientist builds and tests methods for language understanding and generation. They make sure a company's language models work across…

How we wrote this

Role definitions are our own editorial work, checked against occupational classifications, standards, research and employer pages. Titles vary by company, and this is a description of the work, not a job advert.

← Reasoning, planning and searchAll new job families

Start a search

Hiring an Inference-Time Compute Researcher?

Tell us the work. We find the people already doing it, whatever their job title says, and you meet them already interviewed for your role.