The rapid advancement of artificial intelligence is creating an urgent need for better ways to evaluate what AI systems can actually accomplish in the real world. While many existing AI benchmarks measure narrow technical capabilities, the next frontier involves understanding whether advanced AI models and autonomous agents can perform economically valuable, complex, multi-hour professional tasks.

The Mercor Research Fellowship — APEX offers researchers and ambitious domain experts an opportunity to contribute directly to this emerging field.

Through the fellowship, selected participants will receive substantial funding, access to frontier AI models, GPU computing resources, expert human-data support, proprietary datasets, mentorship, and the opportunity to work on research that could influence how the global AI industry measures model capabilities.

The program offers a $40,000 stipend for a three-month fellowship or an $80,000 stipend for a six-month fellowship. Fellows can participate remotely or, where appropriate, work in person from Mercor’s San Francisco office.

Unlike a traditional research fellowship where participants join a predefined research project, the Mercor Research Fellowship is built around the applicant’s own research proposal. Applicants are expected to pitch a specific benchmark, evaluation methodology, or economic study they want to develop.

This makes the fellowship particularly attractive for independent researchers and specialists who already have ambitious ideas about how AI systems should be tested, measured, evaluated, or studied.

The application deadline is September 26, 2026, at 9:00 AM GMT+3.

About Mercor

Mercor is an AI-focused company working at the intersection of human expertise, artificial intelligence, data infrastructure, and the future of work.

The company’s mission is to organize human intelligence to power the AI economy. Mercor operates a platform connecting millions of domain experts with opportunities to contribute their knowledge and expertise toward the development and improvement of frontier AI systems.

According to the program information, experts on Mercor’s platform collectively contribute to training frontier AI models across numerous specialized disciplines.

Mercor is building infrastructure that connects human expertise with advanced AI systems. This includes supporting AI companies and enterprises in understanding how professionals perform complex work and translating real-world expertise into data, workflows, evaluations, and AI agents.

The company also works with enterprise organizations, including large corporations, to capture and understand how highly skilled professionals perform their work.

Mercor’s research activities focus heavily on a critical question:

How can researchers accurately determine whether AI systems are capable of performing meaningful real-world professional work?

Answering this question requires more than conventional AI benchmarks. It requires realistic tasks, expert-designed evaluations, reliable grading systems, and careful research into the economic consequences of AI adoption.

This is where Mercor’s APEX research initiative becomes particularly important.

What Is the Mercor APEX Benchmark Family?

The APEX benchmark family is designed to measure whether frontier AI systems can perform economically valuable work in real professional environments.

Rather than focusing exclusively on short questions, coding puzzles, or standardized academic problems, APEX explores complex and realistic tasks that professionals encounter in their daily work.

These can include:

• Multi-hour agentic tasks in investment banking
• Corporate law assignments
• Professional accounting workflows
• Real-world software engineering problems
• Graduate-level scientific research tasks
• Complex financial workflows
• Long-horizon professional decision-making
• Research-level mathematics and scientific reasoning

The goal is to move AI evaluation closer to the realities of professional work.

For example, an AI system may perform well on a short multiple-choice examination but struggle when required to independently complete a complex workflow involving multiple applications, ambiguous instructions, iterative decision-making, research, documentation, and professional judgment.

Mercor’s APEX benchmarks aim to investigate these differences.

Importantly, the benchmarks are developed and validated using Mercor’s network of domain experts. These experts can include professionals such as:

• Lawyers
• Accountants
• Software engineers
• Scientists
• Consultants
• Finance professionals
• Other specialized practitioners

This expert-driven approach helps ensure that the tasks reflect genuine professional work rather than simplified textbook exercises.

About the Mercor Research Fellowship

The Mercor Research Fellowship funds researchers and domain experts who want to build the next generation of AI benchmarks, evaluation techniques, and economic analyses.

The fellowship is structured around an applicant’s research proposal.

Rather than simply applying to join a general research rotation, candidates are expected to identify an important problem and propose a project they want to pursue.

Applicants may pitch ideas such as:

• A completely new AI benchmark
• A new domain for evaluating AI capabilities
• A more difficult task format
• A new methodology for evaluating AI agents
• A framework for measuring agent reliability
• A method for improving benchmark validity
• A novel grading or scoring methodology
• An empirical study examining AI’s economic impact
• Research investigating how AI affects firms and labor markets
• A study using proprietary data from Mercor’s platform

If selected, fellows receive resources to design, implement, validate, and potentially publish their research.

This includes access to computing infrastructure, expert labor, mentorship, AI models, and research data.

The fellowship therefore provides an opportunity to move from an initial research idea to a completed and validated project.

Research Opportunities in AI Evaluation

One of the major areas supported by the fellowship is AI evaluation.

AI evaluation has become increasingly important as frontier models become more capable. Researchers need reliable methods for determining what models can and cannot do.

However, evaluating advanced AI systems presents several challenges.

A benchmark must answer questions such as:

• Does the task represent a real-world capability?
• Is the benchmark sufficiently difficult?
• Can the benchmark be contaminated by training data?
• Are grading systems reliable?
• Do human experts agree with automated evaluation methods?
• Does high benchmark performance translate into real-world usefulness?
• Can AI agents perform long-horizon tasks reliably?
• How should cost and computational resources be incorporated into evaluation?
• How can researchers distinguish memorization from genuine capability?

The Mercor Research Fellowship encourages fellows to explore these and other important questions.

Successful fellows may design entirely new ways of measuring AI capability.

Research Opportunities in the Economics of AI

In addition to benchmark development, the fellowship supports research examining the economic impact of artificial intelligence.

AI is expected to influence:

• Employment
• Labor markets
• Professional services
• Business productivity
• Organizational structures
• Hiring practices
• Demand for specialized expertise
• Firm strategy
• The structure of work itself

However, measuring these changes requires access to meaningful data.

Mercor’s economics team works to understand how AI is changing the labor market and quantify its broader economic consequences.

The organization has access to unique data sources, including:

• A marketplace involving millions of candidates
• Internal AI usage metrics
• Enterprise data
• Human expertise and matching data
• Information related to the supply and demand for professional expertise

Fellows interested in economics can propose empirical studies using these resources.

Potential projects could investigate how AI changes professional work, how organizations adopt AI, how AI-native firms operate, or how human expertise interacts with increasingly capable AI systems.

Mercor Research Fellowship Program Details

The fellowship is designed to provide flexibility while requiring a substantial commitment from participants.

Duration

The fellowship lasts between:

• Three months, or
• Six months

Admissions are conducted on a rolling basis.

Commitment

Selected fellows must commit to at least:

30 hours per week

Full-time participation is preferred.

This means the program is intended to be a serious research engagement rather than a casual or part-time extracurricular opportunity.

The fellowship may be suitable for individuals who are:

• Taking academic leave
• Completing a summer research project
• Pursuing a flexible stage of a PhD
• Transitioning between professional roles
• Conducting independent research
• Exploring a career in AI evaluation

Location

The fellowship supports remote participation.

Fellows may also have the opportunity to work in person at Mercor’s San Francisco office.

The flexibility of remote participation makes the program potentially accessible to researchers beyond the organization’s physical office locations, although applicants should carefully review the official application requirements.

Admission Model

One of the most distinctive features of the fellowship is its proposal-based structure.

Applicants do not simply apply for a generic research position.

Instead, they are expected to submit a specific idea for:

• A benchmark
• An evaluation methodology
• An empirical research study

The fellowship is funded around the fellow’s proposed project.

Therefore, the quality, originality, relevance, and feasibility of the research pitch are likely to be important factors in the selection process.

What Will Mercor Research Fellows Do?

Selected fellows will participate in the complete research and development process.

This involves much more than theoretical analysis.

1. Propose and Scope a Research Project

The first major responsibility is identifying a meaningful problem that is not adequately addressed by existing research.

Fellows may propose:

• New benchmarks
• Evaluation techniques
• AI measurement methodologies
• Economic studies
• Empirical analyses

A strong project should have a clearly defined scope.

Applicants should ideally demonstrate:

• Why the problem matters
• What existing research is missing
• How the proposed work differs from existing benchmarks
• What methodology will be used
• What resources are required
• What outcomes the project could produce

The ability to transform an ambitious idea into a practical research project will be particularly valuable.

2. Build and Validate AI Benchmarks

For fellows working on benchmark development, the work may involve several stages.

These include:

Designing Task Specifications

Researchers may need to determine what tasks AI models should complete.

Tasks should ideally reflect meaningful and realistic professional challenges.

Developing Grading Rubrics

A benchmark requires reliable ways to evaluate performance.

Fellows may develop detailed rubrics that define:

• What constitutes successful performance
• What common errors look like
• How partial success should be measured
• How human experts should assess outputs

Working With Domain Experts

Mercor provides access to a network of vetted experts.

Depending on the project, fellows may collaborate with professionals such as:

• Lawyers
• Engineers
• Accountants
• Scientists
• Consultants

These experts can help ensure that tasks accurately represent professional workflows.

Piloting Tasks

Before releasing a benchmark, researchers may need to test tasks.

Piloting helps identify problems such as:

• Ambiguous instructions
• Unrealistic assumptions
• Inconsistent grading
• Tasks that are too easy
• Tasks that are excessively difficult
• Poor alignment with real-world work

Calibrating Scoring Systems

Researchers may compare evaluation approaches and refine scoring methodologies.

The objective is to ensure that benchmark scores provide meaningful information about actual AI capability.

Testing Frontier AI Models

Fellows may run advanced AI models against their benchmarks.

This can help researchers identify:

• Where models succeed
• Where models fail
• What types of reasoning remain difficult
• How performance varies across tasks
• Whether models can complete long-horizon workflows

3. Conduct Economic Research

Fellows working on economics may conduct a wide variety of empirical research projects.

Potential activities could include:

• Designing field experiments
• Studying expert behavior
• Analyzing labor market data
• Investigating enterprise AI adoption
• Examining productivity changes
• Studying supply and demand dynamics
• Conducting quantitative analyses
• Investigating organizational transformation

The research may use Mercor’s proprietary datasets, subject to internal review and appropriate access requirements.

4. Publish and Share Research Findings

The fellowship aims to produce tangible research outputs.

Depending on the project, fellows may produce:

• Academic papers
• Research reports
• Open datasets
• New AI benchmarks
• Public leaderboards
• Evaluation methodologies
• Internal research frameworks

Researchers may also collaborate with Mercor’s research and engineering teams to integrate findings into future projects.

This creates an opportunity for fellows to contribute work that could have practical influence beyond the duration of the fellowship.

Mercor Research Fellowship Focus Areas

Mercor has identified several areas of particular interest.

However, applicants are not restricted to these topics.

Strong proposals outside the listed areas are also encouraged.

Long-Horizon Agentic Tasks in Professional Services

AI agents are increasingly expected to perform tasks that involve multiple steps and extended periods of autonomous work.

Mercor is interested in evaluating long-horizon tasks in professional services such as:

• Law
• Finance
• Consulting

This work may extend the organization’s existing APEX-Agents research.

Researchers could investigate how AI systems perform when tasks require:

• Planning
• Research
• Tool usage
• Decision-making
• Iteration
• Error correction
• Long-term context management

Real-World Software Engineering Evaluation

Software engineering evaluation is another major focus area.

Traditional coding benchmarks often focus on isolated programming problems.

However, professional software development involves significantly more complex workflows.

Potential research may explore:

• Codebase navigation
• Software architecture
• Debugging
• Testing
• Documentation
• Multi-file development
• Long-term engineering tasks

This focus area aims to move beyond narrow issue-resolution benchmarks.

Professional Accounting and Finance Workflows

AI systems are increasingly being tested on finance and accounting tasks.

Mercor is interested in developing more realistic evaluations involving professional workflows.

Potential areas could include:

• Financial analysis
• Accounting procedures
• Reporting
• Auditing workflows
• Financial modeling
• Business decision-making

The objective is to understand whether AI can reliably perform economically valuable professional work.

AI for Science Evaluations

AI has the potential to accelerate scientific research.

However, evaluating scientific AI capabilities remains challenging.

The fellowship supports research involving areas such as:

• Research-level mathematics
• Biology
• Materials science
• Theoretical physics

Researchers may investigate how AI systems perform on complex scientific reasoning and research tasks.

New Economic Benchmarks

Some economically valuable skills do not easily fit into conventional AI benchmarks.

Mercor is interested in developing evaluations for capabilities such as:

• Negotiation
• Management
• Professional decision-making
• Other economically significant skills

These areas are difficult to evaluate because performance may depend on context, interaction, and long-term outcomes.

Researchers may therefore develop entirely new benchmark formats.

Novel AI Evaluation Methodologies

Applicants can also focus on improving the science of evaluation itself.

Potential topics include:

Contamination Resistance

Researchers may investigate ways to ensure that AI benchmarks are not compromised by models having previously encountered benchmark data during training.

Rubric Design

A strong evaluation depends on reliable criteria.

Researchers may develop improved methods for creating and validating grading rubrics.

Human-versus-Model Grading Agreement

An important research question is whether automated evaluation systems agree with human experts.

Researchers could study:

• Inter-rater reliability
• Automated grading accuracy
• Human expert disagreement
• Hybrid evaluation systems

Cost-Adjusted Scoring

Different AI systems may require vastly different computational resources.

Researchers may investigate how performance should be measured alongside:

• Computational cost
• API expenses
• Time requirements
• Human intervention

Economics of Human Data Platforms

Mercor is also interested in research examining platforms that connect human expertise with AI development.

Possible topics include:

• Pricing
• Labor matching
• Supply elasticity
• Demand elasticity
• Marketplace dynamics
• Expert incentives

This research could contribute to a deeper understanding of how human intelligence becomes part of the AI economy.

AI Theory of the Firm

Another important research area involves understanding what an AI-native organization looks like.

As AI becomes integrated into organizations, fundamental questions emerge:

• How will firms organize work?
• Which tasks will remain human-led?
• How will AI agents interact with employees?
• How will management structures change?
• How will productivity be measured?
• What new business models will emerge?

Researchers interested in empirical studies of AI-native organizations may find this focus area particularly relevant.

Who Should Apply for the Mercor Research Fellowship?

The fellowship is designed for individuals who are genuinely interested in AI evaluation and research.

Mercor specifically emphasizes that evaluation should not simply be viewed as a temporary stepping stone toward model-building roles.

Successful applicants should demonstrate a genuine intellectual interest in questions surrounding measurement, benchmarking, reliability, and AI capability assessment.

Applicants may have backgrounds in:

• Economics
• Computer Science
• Machine Learning
• Artificial Intelligence
• Statistics
• Measurement
• Psychometrics
• Human-Computer Interaction
• Social Science
• Other adjacent disciplines

The program may also be valuable for domain experts with specialized knowledge.

Bonus experience may include expertise in:

• Agentic AI evaluation
• Reinforcement learning environments
• Law
• Finance
• Medicine
• Scientific research

Key Qualities Mercor Is Looking For

Applicants should demonstrate several important characteristics.

A Genuine Interest in AI Evaluation

Candidates should be interested in the scientific challenge of determining how AI capabilities should be measured.

This involves thinking critically about:

• Benchmark design
• Measurement validity
• Reliability
• Real-world relevance
• Evaluation methodology

A Strong Research Background

Applicants should have relevant academic, technical, or professional experience.

However, interdisciplinary expertise may also be valuable.

For example, a legal professional with strong knowledge of AI could potentially propose a benchmark evaluating complex legal workflows.

Similarly, an economist could propose a study examining how AI changes labor market behavior.

A Specific Research Idea

Perhaps the most important requirement is having a concrete project proposal.

Applicants should be prepared to explain:

• What they want to build or study
• Why it matters
• What gap it addresses
• How they would approach the project
• What resources may be required

A generic interest in AI research alone may not be sufficient.

Ability to Work Independently

The fellowship operates in a fast-paced environment.

Participants should be comfortable with:

• Rapid iteration
• Independent decision-making
• Direct feedback
• Working with real-world research problems
• Managing ambiguity

The program may provide less structured guidance than a traditional academic laboratory.

Significant Time Commitment

Applicants must be able to commit at least 30 hours per week.

Full-time participation is preferred.

Candidates should therefore ensure they have sufficient availability before applying.

Mercor Research Fellowship Compensation

One of the major benefits of the fellowship is its competitive financial support.

Selected fellows can receive either:

• $40,000 for a three-month fellowship, or
• $80,000 for a six-month fellowship

The stipend structure provides significant support for researchers who want to dedicate substantial time to an ambitious project.

The fellowship is therefore positioned as an opportunity for individuals who want to seriously invest in independent AI evaluation or economic research.

Benefits of the Mercor Research Fellowship

In addition to financial compensation, fellows receive access to significant research resources.

1. Unlimited API Credits

Fellows receive access to API credits that can support experimentation with AI systems.

This can be especially valuable for researchers conducting large-scale model evaluations.

2. Dedicated GPU Computing Budget

Selected participants receive resources for GPU compute.

This enables researchers to conduct computationally intensive experiments.

3. Paid Expert and Human-Data Support

Researchers may receive access to paid expert labor and human-data resources.

This is particularly valuable for benchmark development because high-quality professional evaluations often require specialized human expertise.

4. Weekly Individual Mentorship

Fellows receive weekly one-on-one mentorship from members of the APEX research team.

This provides an opportunity to receive feedback and guidance throughout the project.

5. Access to Mercor’s Research Organization

Participants may interact with the broader Mercor research ecosystem.

This can expose fellows to researchers and practitioners working at the intersection of AI, economics, evaluation, and professional expertise.

6. Access to Marketplace and Expert-Network Data

Subject to appropriate review, fellows may access data that can support empirical research.

This creates opportunities for research projects that may not be possible using publicly available datasets alone.

7. Access to Frontier AI Models

Participants may receive access to frontier model APIs.

This enables fellows to test and compare advanced AI systems.

8. Internal Evaluation Infrastructure

Researchers may be able to use Mercor’s internal evaluation tools and infrastructure.

This can significantly reduce the technical barriers involved in building sophisticated benchmarks.

9. Real Enterprise Evaluation Problems

Where appropriate, fellows may gain exposure to real evaluation problems involving enterprise customers.

This can help ensure that research remains connected to practical challenges.

10. Optional San Francisco Workspace

Fellows interested in working in person may have access to a desk at Mercor’s San Francisco office.

11. Professional Research Network

Participants may receive introductions to Mercor’s network of researchers across:

• Frontier AI laboratories
• Academia
• Research organizations

Networking opportunities can be particularly valuable for researchers interested in building long-term careers in AI evaluation.

12. Potential Full-Time Employment Opportunity

Outstanding fellows may be considered for a full-time position on Mercor’s APEX research team after completing the fellowship.

This creates a potential pathway from fellowship participation into a longer-term research role.

How to Develop a Strong Fellowship Proposal

Because the fellowship is built around the applicant’s pitch, prospective applicants should dedicate significant effort to developing a compelling research proposal.

A strong proposal should ideally include the following elements.

Define a Clear Problem

Begin by identifying a meaningful gap.

Ask questions such as:

• What is currently missing from AI evaluation?
• What professional capability is poorly measured?
• What existing benchmark limitation needs to be addressed?
• What economic question lacks sufficient empirical evidence?

Explain Why the Problem Matters

A proposal should demonstrate significance.

Explain how the project could contribute to:

• AI safety
• AI reliability
• Scientific measurement
• Economic understanding
• Professional AI deployment
• Benchmark development

Present a Practical Methodology

Describe how the research will be conducted.

This could include:

• Experimental design
• Data collection
• Expert interviews
• Benchmark construction
• Statistical analysis
• Model testing

Keep the Scope Realistic

The project should be ambitious but achievable within three to six months.

Applicants should avoid proposing a project that is too broad to complete.

Identify Expected Outputs

Consider what the project could produce.

Possible outputs include:

• A research paper
• A benchmark
• An open dataset
• A leaderboard
• An evaluation methodology
• An empirical study

Demonstrate Relevant Expertise

Applicants should explain why they are well-positioned to conduct the proposed work.

Relevant experience could include:

• Previous research
• Technical expertise
• Industry knowledge
• Domain specialization
• Statistical experience
• AI development experience

How to Apply for the Mercor Research Fellowship

Interested applicants should submit their application through the official Mercor application portal.

Applicants should be prepared to present a specific research idea involving:

• A benchmark
• An evaluation methodology
• An empirical study

Because the fellowship is centered around the applicant’s proposal, candidates should carefully refine their project idea before submitting an application.

Applicants should ensure their proposal demonstrates:

1. A clearly defined research problem.
2. The significance of the problem.
3. A practical research methodology.
4. A realistic project scope.
5. Relevant expertise or experience.
6. The potential impact of the proposed work.

Application Links

Applicants can learn more about the fellowship and submit their applications through the official Mercor portal.

APPLY HERE

Important Deadline

The deadline to apply for the Mercor Research Fellowship — APEX is:

September 26, 2026, at 9:00 AM GMT+3

Applicants should avoid waiting until the final hours before submitting their application, particularly because a strong application requires a carefully developed research pitch.

Since the fellowship operates with rolling admissions, submitting a well-prepared application early may also be beneficial.

Why the Mercor Research Fellowship Is a Unique Opportunity

The Mercor Research Fellowship stands out because it combines several elements that are not commonly available together in a single research program.

Participants receive:

• Significant financial support
• Flexibility to propose their own research
• Access to frontier AI systems
• GPU computing resources
• Human expert support
• Individual mentorship
• Proprietary research data
• Real-world enterprise evaluation problems
• Opportunities to publish meaningful work
• Potential pathways toward full-time research positions

Most importantly, the fellowship allows researchers to work on one of the most important emerging questions in artificial intelligence:

How do we know whether increasingly powerful AI systems can perform meaningful, economically valuable work in the real world?

As AI models move beyond answering questions and generating text toward operating as autonomous agents, traditional evaluation methods may become increasingly insufficient.

The need for better benchmarks, stronger evaluation methodologies, and rigorous economic research is therefore likely to grow significantly.

Researchers who want to contribute to this emerging field have an opportunity to develop ambitious projects with substantial institutional and technical support.

Visit HERE to learn more about the Mercor Research Fellowship APEX 2026.

Conclusion

The Mercor Research Fellowship — APEX 2026 presents a highly competitive opportunity for researchers, economists, AI practitioners, computer scientists, statisticians, domain experts, and interdisciplinary thinkers interested in the future of artificial intelligence evaluation and the economics of AI.

With funding of up to $80,000, access to advanced computing infrastructure, frontier AI models, expert networks, proprietary data, and direct mentorship from the APEX research team, the fellowship provides selected participants with the resources needed to transform an ambitious research idea into a meaningful project.

The program’s proposal-driven model makes it particularly suitable for individuals who already have a clear idea about a benchmark they want to build, an evaluation challenge they want to solve, or an economic question about AI they want to investigate.

Whether the proposed work focuses on long-horizon AI agents, software engineering, law, finance, accounting, scientific reasoning, AI evaluation methodology, labor markets, or the future structure of AI-native firms, Mercor is looking for strong ideas that can advance understanding of AI’s real-world capabilities and economic consequences.

For researchers seeking to work at the intersection of artificial intelligence, measurement, professional expertise, and economic transformation, the Mercor Research Fellowship offers an opportunity to conduct impactful work while receiving substantial financial and technical support.

Interested candidates should develop a focused and compelling research proposal and submit their applications before the deadline of September 26, 2026, at 9:00 AM GMT+3.

For more global Fellowship opportunities visit OFY HERE