About the role
The reasoning team studies how models think. That covers chain-of-thought, test-time compute, search over candidate solutions and self-verification, and the training that makes extra thinking worth the tokens. The aim is concrete: models that get hard problems right, notice their own errors and know when they are unsure.
Reasoning capability is what lets Clexerl Go's agents plan across many steps without losing the thread, and it shapes how the Memory and Agents teams design their systems. The team also supplies the hardest problems to Evaluations, because a reasoning claim that cannot be tested on held-out, uncontaminated tasks does not count.
Research here alternates between close reading of model traces and large controlled experiments. Scientists are expected to understand what a chain of thought actually does to the answer, including when it is post-hoc rationalisation, and to design experiments that tell the two apart.
What you’ll do
- Develop methods that scale test-time compute, including long chains of thought, best-of-n sampling, tree search and learned verifiers.
- Train models to verify and correct their own reasoning, and measure whether the corrections are real or cosmetic.
- Study the faithfulness of chains of thought by intervening on intermediate steps and observing how answers change.
- Work with the reinforcement learning team on reward signals for reasoning, including outcome and process supervision.
- Build hard, contamination-resistant problem sets in mathematics, code and scientific reasoning.
- Analyse failure modes such as overthinking, loops and confident wrong answers, and propose training interventions.
What you’ll bring
- A research record in reasoning, language models, program synthesis or a closely related area, with work you can discuss in technical depth.
- Strong grasp of how language models are trained and sampled, including temperature, nucleus sampling and the effect of length on accuracy.
- Experience with reinforcement learning or search methods applied to sequence models.
- Mathematical maturity: you can read a proof or a derivation and notice where a claim does not follow.
- Rigorous experimental design, including held-out splits, contamination checks and confidence intervals.
- Curiosity about cognition and a willingness to be told by the data that your favourite theory is wrong.
Nice to have
- Experience with formal verification, theorem provers or proof assistants.
- Research on learned verifiers, critics or self-consistency methods.
- Background in competition mathematics or scientific problem sets as a source of evaluation tasks.
- Familiarity with efficient inference for long outputs, such as speculative decoding.
Work you might take on
- 01
Train a verifier that predicts correctness from a partial chain of thought and use it to prune search early.
- 02
Measure how often edits to a model's stated reasoning change its final answer, and what that says about faithfulness.
- 03
Build a held-out set of multi-step quantitative problems with automatically checkable answers and no public leakage.
- 04
Study overthinking by finding the point where extra tokens stop helping and training the model to stop there.
Compensation
$340K – $560K
The US base salary range for this role. Every offer also includes equity in Clexerl and, when it spins out, in the venture you help build. We tell you the reasoning behind your number.
Benefits
Health, fully covered
Medical, dental and vision for you and your dependants, premiums paid in full.
Two kinds of equity
Equity in Clexerl, plus a stake in the venture you help build when it spins out.
22 weeks parental leave
Fully paid, for every parent, however your family arrives.
Rest that sticks
Flexible time off with a four-week minimum we actually check on, plus a company-wide week in December.
Learning and compute
A yearly budget for courses, conferences and books, and compute credits for side projects.
401(k) with a 4% match
Matched from your first paycheque and vested immediately.
How we hire
Apply 5 minutes
Send links to what you have built and a few lines on why this role. No cover letter needed.
Intro call 30 minutes
A conversation with the hiring manager about the work, the team and what you want next.
Work sample 2–4 hours
A paid take-home or a pairing session on a real-shaped problem. You pick the format.
Studio day One day
Meet the team, go deep on your work and ours. Remote or in person.
Offer Within 5 days
A written offer with the full range, the equity and the reasoning behind both.
Opens an email with a short template. Links beat résumés.