Mercor AI Research Fund Grants: Benchmarks

Mercor's mission is to organize human intelligence to power the AI economy. We're a leading AI data company, building the layer between human expertise and frontier models. Millions of domain experts on the platform are paid over $4 million per day to train frontier AI models. Mercor's APEX benchmark family measures AI's real-world impact on professional work. Mercor Enterprise brings this same infrastructure to Fortune 500 companies: helping companies capture how their best people actually work, translating that expertise directly back into agents.

Mercor is creating a new category of work where expertise powers AI advancement. Achieving this requires an ambitious, fast-paced and deeply committed team. You’ll work alongside researchers, operators, and AI companies at the forefront of shaping the systems that are redefining society. Mercor is a profitable Series C company valued at $10 billion. We work in-person five days a week in our San Francisco, NYC, or London offices.

Mercor AI Research Fund Grants: Benchmarks

One of the biggest challenges facing the industry today is determining whether frontier AI is capable enough to deliver value in real-world settings. A model can perform well on benchmarks and still struggle in production. That’s why investing in research, realistic evaluations, and robust verification is critical.

Mercor is seeking to fund AI research on benchmarking. This is separate from the Mercor Research Fellowship, which funds individual experts (apply HERE), our $5m of safety funding awards (apply HERE), and our research funding on the economic impacts of AI.

The grant supports:

• The time of experts from Mercor's platform

• Researcher hours

• API costs

• Stipends for event and conference attendance

Budget Categories

Applications are grouped into three budget tiers. Please state your tier clearly in your application and ensure your overall budget ask falls within it:

• Tier 1: under $50,000: Suited to focused, well-scoped projects such as pilot studies, evaluations, replications, or extensions of existing work. These applications undergo a streamlined review.

• Tier 2: under $200,000: Suited to substantial projects requiring dedicated personnel time, meaningful compute, or data collection. Reviewed with proportionally greater attention to the budget breakdown and implementation plan.

• Tier 3: under $500,000: Reserved for ambitious, multi-component research programs. These applications receive the highest level of scrutiny: reviewers will expect a detailed justification of each spend category, a clear account of team capacity to execute at this scale, and milestones against which progress can be assessed.

The level of review scrutiny scales with the size of the ask, so applicants should request the smallest budget that credibly supports the proposed work.

Research Agenda

We are interested in submissions in the following area:

• Real-world evaluations: measuring whether benchmark performance translates into reliable performance on realistic tasks and workflows

• Environments: building high-fidelity, interactive environments that capture the complexity of real-world work

• Long-horizon tasks: evaluating and improving models’ ability to plan, execute, and adapt across extended workflows

• Agentic capabilities: tool use, coordination, memory, and autonomous execution

• Reasoning and problem-solving: improving performance on complex, ambiguous, or underspecified tasks

• Post-training: developing better data, rewards, and training methods to improve model capabilities

• Evaluation methodology: creating more robust measures of capability, reliability, and real-world utility

Eligibility

• Researchers at universities and non-profit research organizations, as well as public benefit corporations and small startups (<20 people) that do not sell AI training data. Awards are made to the researcher’s institution.

• Applicants must be based in the United States or the United Kingdom

We expect grantees to publish a paper, an open dataset, a public methodology, or a tool the field can use, released under an open license that permits commercial use.

Timeline

• Application deadline: November 27, 2026, 11:59 PM PT (UTC−8)

• Notification to applicants: by December 18, 2026

Late or incomplete applications will not be reviewed. All applicants will be notified of the outcome whether or not they are shortlisted.

How to apply

We invite interested researchers to submit a brief Expression of Interest (EOI) before preparing a full proposal. The EOI should be ~1–3 pages and include:

• Research question and background (max 250 words): What are you trying to do? State your objective in plain language, with minimal jargon. How is this problem addressed today, and what are the limits of current practice? Who cares: if you succeed, what difference will it make, and to whom?

• Project proposal (max 400 words): Concrete technical and implementation detail: your approach, methods, datasets, models, compute, and evaluation plan.

• Novelty (max 200 words): What is new in your approach relative to existing and ongoing work, and why do you believe it will succeed where current practice falls short?

• Risks (max 150 words): What are the main technical and execution risks, and how will you mitigate or detect them early?

• Team (max 100 words; if the team is large, up to 200): The researchers on the project and their relevant background and prior work.

• Proposed outcomes and timeline: Expected deliverables (papers, artifacts, datasets, tools) and key milestones, including how long the work will take. Include the mid-term and final “exams”: the checkpoints and success criteria by which we will know whether the project is on track and succeeds.

• Budget overview: How much will it cost? Provide a specific overall budget ask, your budget tier (see Budget Categories), and a breakdown by spend category (e.g., personnel, compute, data, travel/other).

  • Other sources of funding for this project should be clearly identified in the proposal budget

Guidance for applicants.

• Strong EOIs are specific rather than high-level: they name the methods, systems, and evaluation criteria they will use, and they engage directly with existing literature and ongoing efforts rather than describing the problem space in general terms.

• The EOI should briefly establish the PI’s (and key team members’) relevant background and prior work; proposals without demonstrated expertise in the research area are rarely competitive. Researchers with experience with agentic evaluation, RL environments, and post-training are encouraged to apply

• Proposals that duplicate work already being pursued by many other groups are unlikely to advance unless they offer a clearly differentiated angle.

• Please do not include confidential or proprietary information in your Expression of Interest. Submitting an Expression of Interest does not obligate Mercor to make an award.

Award terms

Selected grantees will sign Mercor’s research award agreement. Key terms include:

• Grantees keep ownership of their work and grant Mercor a non-exclusive license to use it, including commercially.

• Mercor may identify grantees and their institutions as award recipients, including by name and logo. Grantees may use Mercor’s logo with Mercor’s approval.

• For benchmark and evaluation projects, Mercor receives early access to the benchmark before public release and the exclusive right to build and host a private held-out test set, at Mercor’s cost.

• Grantees agree not to work with specified Mercor competitors on substantially similar benchmarks or evaluations during the award and for 12 months afterward.

• Grantees are free to publish their findings. Mercor reviews drafts in advance only to remove its confidential information.

• Awards are subject to tax documentation, sanctions screening, and periodic progress reports.

About Mercor

Mercor's mission is to organize human intelligence to power the AI economy. We're a leading AI data company, building the layer between human expertise and frontier models. Millions of domain experts on the platform are paid over $4 million per day to train frontier AI models. Mercor's APEX benchmark family measures AI's real-world impact on professional work. Mercor Enterprise brings this same infrastructure to Fortune 500 companies: helping companies capture how their best people actually work, translating that expertise directly back into agents. Mercor is a profitable Series C company valued at $10 billion.