NuminaMath wins the first AIMO Progress Prize with an open tool-using math model

The winning system fine-tuned DeepSeekMath-Base 7B and combined chain-of-thought with Python tool execution, self-consistency and internal validation sets.

What happened?

NuminaMath-7B-TIR won the first AIMO Progress Prize by solving 29 of 50 problems on the private test set.

Why it matters

It provided an open and reproducible counterpoint to closed frontier math systems and helped popularize tool-integrated math reasoning.

Significance rationale: NuminaMath showed that a relatively small open model plus tool-integrated reasoning and carefully curated math data could win a public math-AI competition.

Historical context

Primary evidence

Evidence & status at a glance

Significance
H2 · Field milestone
Verification
Partially verified
Evidence level
E3 · Inspectable artifact
AI role
AI primary, human verified
Formal assurance
Artifact available

How should these labels be read? →

Contribution & verification

Mathematical novelty
Benchmark result
Interface
Informal, Numeric, Hybrid
AI contribution
The model generates reasoning and invokes Python during inference to solve competition problems.
Human contribution
The Numina and Hugging Face teams curated training data, designed TIR decoding and tuned the competition system.
Verification
Partially verified · E3 · Inspectable artifact · Artifact available
Last reviewed
2026-09-11

Technical details

Systems
numinamath
Organizations
numina, hugging-face
Problems
aimo-progress-prize
Methods
tool-integrated-reasoning, self-consistency, fine-tuning

Sources & artifacts

Verification history

  • 2024-07-11 · Partially verified · E2 · Research record

    The Chronicle records this event from the cited primary or official source; the status does not imply independent Chronicle verification.

  • 2026-09-11 · Partially verified · E3 · Inspectable artifact

    This corpus refresh added a directly inspectable public artifact maintained by the authors or releasing organization. E3 records artifact-level inspectability and does not imply independent verification of the underlying scientific or mathematical claim.

Corrections

No corrections recorded.