OpenReasoning-Nemotron
lastmod 2025-07-21created 2025-07-211 edits
OpenReasoning-Nemotron 32B/14B/7B/1.5B models released
About
- Derivative models of Qwen2.5-32B-Instruct (a reasoning model), specialized in reasoning for solving math, code, and science problems
- Can be used for commercial and non-commercial research purposes
- Supports a context length of 64,000 tokens and comes in 1.5B, 7B, 14B, and 32B sizes
- Achieves best-in-class reasoning performance on a range of benchmarks including LiveCodeBench, GPQA, and MMLU-PRO
- When reasoning with multiple agents combined (GenSelect), achieves even better performance on math, code, and science benchmarks than a single model
- GenSelect: runs multiple model reasoning passes in parallel, then selects the best solution