NVIDIA Nemotron 3: Efficient and Open Intelligence

Benchmark Model Rank Results
code-generation-on-livecodebench-v5Nemotron 3 Super#2Pass@1: 81.19
agents-on-bench-airlineNemotron 3 Super#7Accuracy: 56.25
agents-on-bench-retailNemotron 3 Super#8Accuracy: 62.83
code-generation-on-scicodeNemotron 3 Super#12Accuracy: 42.05
instruction-following-on-ifbenchNemotron 3 Super#16Accuracy: 72.56
agents-on-benchNemotron 3 Super#24Accuracy: 61.15
multi-task-language-understanding-on-mmlu-proNemotron 3 Super#27Accuracy: 83.73
code-generation-on-swe-bench-multilingualNemotron 3 Super (OpenHands)#35Accuracy: 45.78
deep-research-agents-on-browsecompNemotron 3 Super#53Accuracy: 31.28
code-generation-on-swe-bench-verifiedNemotron 3 Super (OpenHands)#63Accuracy: 60.47