OpenMath-Nemotron
NVIDIA's open math-reasoning family (1.5B–32B) that won AIMO-2 — the 14B powered the first-place Kaggle solution.
About
OpenMath-Nemotron is NVIDIA's entry in the open reasoning race: four Qwen2.5-based models from 1.5B to 32B parameters, trained purely with supervised fine-tuning on millions of NeMo-Skills-generated math, science and code solutions. They ship in two flavors, plain chain-of-thought and tool-integrated reasoning, plus an optional GenSelect mode that generates many solutions in parallel and picks the best one.
The family earned its reputation in competition: a version of the 14B model took first place in the AIMO-2 Kaggle contest, and the 32B posts 78.4% on AIME 2024 in tool-integrated mode. Everything is out in the open under CC-BY-4.0, which makes these some of the cleanest starting points available for anyone wanting to do their own RL experiments on a strong reasoning base.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
- Vendor-reported (HF model card): OpenMath-Nemotron-32B TIR 78.4% AIME24 / 64.2% AIME25; CoT 76.5% / 62.5%; with GenSelect up to 93.3% / 80.0%
- OpenMath-Nemotron-14B powered the 1st-place solution in the AIMO-2 Kaggle competition
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 9/10
Scored 9/10 (TIER verified, metrics-v1). Evidence: NVIDIA release with full model cards (nvidia/OpenMath-Nemotron-32B etc.) and detailed benchmark tables; 1st place in AIMO-2 Kaggle; CC-BY-4.0 license confirmed via HF API; open dataset and NeMo-Skills pipeline. Strengths: fully open weights and data recipe, permissive license, competition-validated. Docked for: (1) headline benchmark numbers are vendor-reported. License: CC-BY-4.0.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 8.0/10
- Maintenance · recency, releases, issue load
- 8.0/10
- Openness · license + docs
- 9.0/10
- Manual review · sanity read + red-flag check
- 9.0/10
- · NVIDIA; 1.5B/7B/14B/32B family; AIMO-2 winning lineage
- · 32B: 78.4% AIME24 TIR; 14B powered 1st-place AIMO-2 Kaggle solution
- · Benchmarks are vendor-reported