The gold standard for mathematical AI
Fields Model Initiative
Training LLMs for mathematics that converge to Fields Medal performance. A non-profit initiative.
Training LLMs for mathematics that converge to Fields Medal performance. A non-profit initiative.
The Fields Model Initiative accelerates open-source research on mathematical LLMs by providing large-scale compute (128+ H100 GPUs) to vetted projects in exchange for open mathematical data and open-source artifacts. Researchers apply with a proposal, contribute high-quality math datasets under an open license, and, if accepted, use our GPUs to train models whose outputs are shared with the community.
Researchers apply with a proposal for compute. This can range from finetuning LLMs on mathematical data, to analysing LLM checkpoints to understand reasoning better. We vet your proposal and decide whether to give access.
As a "fee" to get access to the compute offered by the Fields Model Initiative, you need to provide data points in mathematics as directed by us, under an open licence.
We provide you access to our GPUs. You make your artefact, which will benefit mathematicians, public under an open-source licence.
Our long-term goal is a fully open-source LLM that can genuinely assist mathematicians across all domains, ultimately approaching Fields Medal–level capability.
One of our primary aims of improving the state of the art of LLM capability is by making available compute grants. LLMC, NII provides grant support through compute access and related engineering efforts. These grants are intended to support researchers carrying out large training runs and experiments to devise better models in AI4Math.
To improve the state of mathematical benchmarks and datasets, which is skewed towards specific types of datasets (e.g. proof generation), we ask researchers who are given access to compute to contribute a small amount of mathematical datapoints, in order to improve, over time, the state of mathematical benchmarking and datasets. This requirement will be waived for large partners.
AIMO Proof Pilot focused on assessing whether fully open LLMs (the Olmo family, the SmolLM family, and Apertus), each of which has detailed training mixes information and checkpoints released, can be adapted and posttrained to output proof at IMO level. The competition ran for one month, with invited teams competing, and the final eval took place on June 25th, 2026. After the models were uploaded to Kaggle, six IMO-level problems were served and their proof outputs returned. These were graded by AIMO on a markscheme published in advance by a team of graders with IMO experience, with at least two graders per proof. Each competing team from the AIMO Proof Pilot was given access to up to 24 H200 GPUs through LLMC, NII to train their models, as well as engineering assistance, throughout May 2026 to mid June 2026. The data fee was waived to allow contestants to fully focus on creating the best-possible models.
The AIMO Interpretability Challengeis a competition accepted at NeurIPS 2026 that focuses on distinguishing robust from spurious reasoning in LLMs, as a strong final-answer accuracy may not reveal whether a model relies on stable reasoning mechanisms or exploits brittle reasoning shortcuts. Competitors need to upload a system that assesses whether a given model solves a given problem reliably. Building on AI Mathematical Olympiad (AIMO) problems and submissions, together with resources from the Fields Model Initiative, the competition will provide (1) newly-published olympiad-level math reasoning problems and their symbolic representations, allowing generation of novel functional variants, (2) access to frontier reasoning models, and (3) assessments of models' adversarial robustness on these problems. Contestants are supported by compute from the LLMC, NII through the Fields Model Initiative.
AIMO Proof Pilot focused on assessing whether fully open LLMs (the Olmo family, the SmolLM family, and Apertus), each of which has detailed training mixes information and checkpoints released, can be adapted and posttrained to output proof at IMO level. The competition ran for one month, with invited teams competing, and the final eval took place on June 25th, 2026. After the models were uploaded to Kaggle, six IMO-level problems were served and their proof outputs returned. These were graded by AIMO on a markscheme published in advance by a team of graders with IMO experience, with at least two graders per proof. Each competing team from the AIMO Proof Pilot was given access to up to 24 H200 GPUs through LLMC, NII to train their models, as well as engineering assistance, throughout May 2026 to mid June 2026. The data fee was waived to allow contestants to fully focus on creating the best-possible models.
The AIMO Interpretability Challengeis a competition accepted at NeurIPS 2026 that focuses on distinguishing robust from spurious reasoning in LLMs, as a strong final-answer accuracy may not reveal whether a model relies on stable reasoning mechanisms or exploits brittle reasoning shortcuts. Competitors need to upload a system that assesses whether a given model solves a given problem reliably. Building on AI Mathematical Olympiad (AIMO) problems and submissions, together with resources from the Fields Model Initiative, the competition will provide (1) newly-published olympiad-level math reasoning problems and their symbolic representations, allowing generation of novel functional variants, (2) access to frontier reasoning models, and (3) assessments of models' adversarial robustness on these problems. Contestants are supported by compute from the LLMC, NII through the Fields Model Initiative.
Pontus Stenetorp's group and LLMC, NII provide access to compute resources and related engineering support for grant recipients.
All applications will be reviewed by our team and must be made in a specified format. This is to ensure that access to hardware resources is as smooth as possible and no exciting idea is left behind. You must also agree that the data artefact you produce will be available under an open-source licence.
We have a standardized process for submissions.