1,000+ Opportunities
Find the right grant
Search federal, foundation, and corporate grants with AI — or browse by agency, topic, and state.
"AIMO Interpretability Challenge at NeurIPS 2026 for Detecting Robust versus Spurious Reasoning in Math Olympiad Language Models" is currently closed and not accepting applications.
The AIMO Interpretability Challenge, part of the NeurIPS 2026 competition program, asks whether interpretability or behavioral methods can tell when a language model genuinely solves a problem and when its reasoning is spurious. Given an unseen olympiad math problem and an AIMO 3 model, a system must label the model's solution as robust or spurious, and entries are ranked by accuracy.
Methods can use white-box signals such as weights, activations and chain-of-thought, or black-box analysis. There are two tracks: a Main Track with no model size limit, and a Small Models Track for models under 10B parameters. The models include Qwen3.
5-4B, Olmo-3-7B-Think, DeepSeek-R1-0528-Qwen3-8B and gpt-oss-120b. The $12,500 prize pool pays $2,500, $1,000 and $500 in the Main Track and $1,500, $700 and $300 in the Small Models Track. Each track also has a $1,000 White-Box Award, and $4,000 goes to writeup prizes.
Submissions are offline Codabench bundles with a 3,600-second limit. Final submissions are due 1 November 2026, technical reports by 15 November, and the workshop is on 12 December 2026 in Paris.
Get a weekly digest of new grants like this
A free weekly digest of new foundation and federal funding opportunities as they're added to Granted. Unsubscribe anytime.
Or search similar grants →According to the current listing, eligibility includes: Open to academic, independent and industry participants with no limits on team size, affiliation or career level; AIMO participation is not required. Prizes cannot be paid to sanctioned persons or recipients in North Korea, Russia, Belarus, Iran, Cuba or Syria. Confirm the full requirements in the official notice before applying.
The current listing shows $12,500 total prize pool: Main Track $2,500/$1,000/$500; Small Models Track $1,500/$700/$300; $1,000 White-Box Award per track; $4,000 in writeup prizes. Verify award ceilings, matching requirements, and allowable costs in the official notice.
The published deadline was November 1, 2026, which has passed. Check the official notice for any future application windows before investing time in a proposal.
AIMO Interpretability Challenge at NeurIPS 2026 for Detecting Robust versus Spurious Reasoning in Math Olympiad Language Models is funded by AIMO Interpretability Challenge organizers (researchers from NII Japan, LMU Munich, University of Tübingen, Oxford, UCL, Masaryk University and KInIT). Verify program details on the funder's official page before applying.
Start from the official opportunity page linked in this listing — it carries the sponsor's submission instructions.
The Farm Robotics Challenge is a student competition rather than a research grant, and the distinction determines who should spend time on it: the money flows to teams of enrolled students working under a faculty or teacher advisor, not to principal investigators, and the deliverable is a working robotic system demonstrated in an agricultural setting rather than a paper. Organized by UC ANR Innovate with the NSF-USDA AI Institute for Next Generation Food Systems and backed by California's Jobs First initiative, the challenge asks teams to build autonomous systems or robotic attachments addressing real production tasks - planting, weeding, harvesting, pest management, field mapping and data collection. Total prize funding exceeds $100,000, anchored by a $50,000 Grand Prize presented by The Reservoir; in a recent cycle that grand prize was structured as a SAFE investment rather than a straight cash award, which matters because it implies an expectation that the winning team may incorporate, and teams uninterested in commercialization should confirm the current cycle's award instrument before entering. Three divisions keep the competition tractable across education levels: Division I for four-year universities with a cap of 50 percent graduate students per team, Division II for community colleges and trade schools, and Division III for middle and high school teams in grades 7 through 12. Every team requires an advisor, typically faculty, staff, a graduate student or a teacher. The cycle runs long, which is the main planning consideration - pre-registration opens August 3, 2026, project proposals are due November 13, 2026, final projects are due May 2, 2027 and awards are presented in June 2027, meaning roughly nine months of sustained student effort across two academic terms. Scale is real: a recent edition drew 96 teams from 13 countries. This is a good fit for agricultural engineering, robotics and CS programs seeking a structured capstone with industry exposure, and a poor fit for anyone looking for research funding on a normal grant timeline.
MIDAS at the University of Michigan is recruiting its 2027 cohort of Michigan Data Science Fellows, a two-year postdoctoral program. It trains fellows to apply AI and data science to social science, biomedical and healthcare research, and to develop data science methodology. Fellows work in an interdisciplinary environment with other MIDAS fellows and more than 750 MIDAS affiliate faculty. The program is recruited alongside the Eric and Wendy Schmidt AI in Science Postdoctoral Fellowship, but it is the only MIDAS track that explicitly covers biomedical and healthcare applications. Applications are due November 23, 2026 through the MIDAS portal (myumi.ch/dgbJV). Advertised salaries range from $70,000 to $86,000 per year.
ClimX is a machine learning challenge to build fast emulators of the NorESM2-MM Earth System Model, judged on how well they reproduce climate extremes rather than mean climate. It is run by the Image and Signal Processing Lab at the University of Valencia, and ESA Phi-lab provides the prizes. Models predict daily global fields at about 1 degree resolution for seven surface variables: mean, maximum and minimum temperature, precipitation, humidity, sea-level pressure and wind speed. Training data covers the historical period from 1850 to 2014 and the SSP1-2.6, SSP3-7.0 and SSP5-8.5 scenarios to 2100. SSP2-4.5 is held out for testing. Rankings use 15 ETCCDI-style extreme indices, scored by region-wise normalized Nash-Sutcliffe efficiency across AR6 land regions. An optional probabilistic track scores uncertainty with CRPS. The main track pays EUR 1,800, 1,000 and 500 to the top three. Finalists must run inference on a single accelerator of 80 GB or less within 24 hours. Submissions close on 15 December 2026, and top teams may be invited to co-author the competition paper.
DARPA and NSF launched a joint program on June 1 to fund university work on AI interpretability, control, and adversarial robustness. Awards run $750K to $3M+ per project, the forum launches this summer, and the universities listed in the AI Forge repository will sit closest to the money. The Request for Information closes June 22.
Read articleNVIDIA committed $250 million over five years to UK agentic safety and security research, delivered as compute access through UK-based Cloud Partners. No open call, no cash, and the commercial terms are not yet agreed. A breakdown of compute-as-grant and how to budget for it.
Read articleThe RFI closed June 22. Now the forum is standing up. AI Forge is a jointly governed, university-led venture funding interpretability, control, and adversarial robustness in one-year Project Ventures — here's how the 15 challenges, the CAISI tie, and the nonprofit administrator reshape who gets funded.
Read article