ABOUT THINKING MACHINES
The mission of Thinking Machines is to build AI that extends human will and judgment. We are training frontier models with Inkling, developing Tinker to let people make models their own, and crafting interfaces that broaden human-AI communication. We believe the future worth building is human, and we're hiring people who want to build it.
ABOUT THE ROLE
As a Research Product Manager at Thinking Machines, you’ll partner closely with researchers and engineers to shape the capabilities of our frontier models and the methods that let people make those models their own. You’ll connect research direction, experimental evidence, and real-world needs, helping the team decide which capabilities to pursue, what to learn next, and where to invest.
A central focus of this role is mid-training: the work between pre-training and post-training that shapes what a model knows, the capabilities it develops, and how effectively it can learn afterward. This includes data and synthetic data strategies, training recipes, context extension, and capabilities across coding, math, reasoning, and knowledge-intensive work. You’ll help connect these choices to downstream model quality, adaptability, and usefulness.
The most interesting questions often span training stages. Which capabilities should we build into the model’s foundation? What makes a model easier to specialize through further training? How do we tell whether an improvement transfers to unfamiliar tasks, survives later training, and matters to the people using the model? You’ll help turn these questions into focused research priorities and clear decisions.
This role is for someone who enjoys working deeply inside research, can move between technical details and the broader opportunity, and brings independent judgment to problems where the right goal or metric is still emerging. Success means helping the team build differentiated technology and turn scientific progress into useful capabilities.
WHAT YOU’LL DO
- Partner with research leaders to shape capability priorities, sequence work, and connect mid-training efforts to the broader model and customization agenda.
- Help translate capability goals into concrete research questions across data sourcing, curation, synthetic data generation, mixture design, and training recipes. Work with researchers to identify the experiments that will most meaningfully inform the next decision.
- Connect evaluations to the capabilities we actually want to build. Help distinguish benchmark improvements from generalizable gains, and assess what persists through downstream training and adaptation.
- Bring evidence from users, developers, and partners into research direction: where models lack knowledge, struggle with long context, fail to generalize, or require too much data or compute to specialize.
- Help define what makes a strong foundation for further learning, including how mid-training choices affect downstream training stability, sample efficiency, capability retention, and customization quality.
- Maintain a clear view across research, data, evaluations, training infrastructure, and product dependencies. Surface tradeoffs around capability, compute, data quality, and iteration speed, and bring the right people together to resolve them.
- Create lightweight plans and decision records that keep ambitious work moving while leaving room for discovery. Make ownership, dependencies, evidence, and unresolved questions clear.
- Help carry promising research into broader use through internal adoption, model evaluation, product integration, and feedback from real applications.
SKILLS AND QUALIFICATIONS
Minimum qualifications:
- Experience driving technically demanding products or research initiatives in an AI lab, research organization, technical startup, or similarly ambiguous environment.
- Experience working closely with model training, training data, evaluations, customization, or ML infrastructure, with enough technical depth to engage substantively with researchers and engineers.
- Strong judgment about experimental evidence: the ability to clarify assumptions, question whether a metric captures the intended outcome, and help teams decide what a result should change.
- A track record of turning open-ended problems into focused work across teams, including identifying missing connections and taking ownership where responsibilities are unclear.
- Clear written and verbal communication, intellectual honesty, and the ability to earn trust through curiosity, sound reasoning, and follow-through.
Preferred qualifications (we encourage you to apply if you meet some but not all of these):
- Experience with continued pre-training, mid-training, supervised fine-tuning, reinforcement learning, or domain adaptation.
- Familiarity with synthetic data pipelines, data quality systems, training mixture design, or evaluations for coding, reasoning, long-context understanding, and knowledge acquisition.
- Experience connecting model development to downstream adaptation, including understanding how training choices affect a model’s ability to learn new tasks and domains.
- Experience helping turn a novel research capability into a useful product, platform, or repeatable method.
- A background as a technical founder, engineer, researcher, or applied scientist who moved into product, or as a product leader working closely with frontier model development.
LOGISTICS
- Location: This role is based in San Francisco, CA.
- Compensation: Depending on background, skills and experience, the expected annual salary range for this position is $350,000 - $400,000 USD.
- Visa sponsorship: We sponsor visas. While we can't guarantee success for every candidate or role, if you're the right fit, we're committed to working through the visa process together.
- Benefits: Thinking Machines offers generous health, dental, and vision benefits, unlimited PTO, paid parental leave, and relocation support as needed.