Senior Research Scientist, Model Evaluation (Remote‑Flexible)

SupportFinity™; New York (NY); On-site; USD 140,000 - 200,000; Full time

Benefits offered by this job

Open and inclusive culture

Work on cutting-edge AI research

Weekly lunches and snacks

Full health and dental benefits

Parental Leave top-up

Enrichment perks for arts and wellness

Remote-flexible, offices in multiple'

Co-working stipend

Six weeks of vacation

Job summary

Cohere is seeking a Senior Research Scientist, Model Evaluation in New York to lead the development of next-generation evaluation methods and infrastructure for measuring LLM progress. You will design ambitious benchmarks and work across teams to translate model feedback into robust, repeatable assessments.

The role emphasizes research, evaluation efficiency, and building scalable tooling to probe model performance, with a focus on aligning measurements with high-value capabilities.

Qualifications

• Experience creating evaluation benchmarks for AI/ML models.
• Experience translating model feedback into repeatable evaluations.
• Research to advance state-of-the-art in LLM evaluation methods.
• Strong software engineering skills.

Responsibilities

• Create ambitious new evaluation benchmarks that push model capabilities.
• Collaborate with cross-functional teams to translate feedback into trustworthy evaluations.
• Conduct research to advance LLM evaluation methods and improve efficiency.
• Build scalable, reusable tools for analyzing model performance.

Skills

Evaluation benchmarks

LLM training

LLM data synthesis

Evaluation tooling

Software engineering

Tools

Data synthesis pipelines

Scalable tooling

Job description

Cohere is seeking a Senior Research Scientist, Model Evaluation in New York to lead the development of next-generation evaluation methods and infrastructure for measuring LLM progress. You will design ambitious benchmarks and work across teams to translate model feedback into robust, repeatable assessments.

The role emphasizes research, evaluation efficiency, and building scalable tooling to probe model performance, with a focus on aligning measurements with high-value capabilities.