Model profile
Nemotron 3 Super 120B A12B
Evidence summary
Safety. Nemotron 3 Super 120B A12B has an estimated Safety rank of #196; its 90% source-sensitivity interval is #107–#279. Its behavior-only rank is #197; company governance moves the combined estimate to #196. Published Safety evidence spans 11 eval lineages and 6 of 7 components. Its strongest relative result is Enkrypt AI Safety Leaderboard (harmful_attack_non_success_rate, #63 of 270); its weakest is Pokee-Isaac model card — DTAP (benign_task_success_rate, #6 of 6).
Freedom. Nemotron 3 Super 120B A12B has an estimated Freedom rank of #533; its 90% source-sensitivity interval is #336–#634. Published Freedom evidence spans 5 eval lineages and 1 of 1 components. Its strongest relative result is Enkrypt AI Safety Leaderboard (cbrn_attack_non_success_rate, #13 of 270); its weakest is SM-Bench (overfit, #89 of 92).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Artificial Analysis ↗nvidia-nemotron-3-super-120b-a12b
- NVIDIA ↗qc69jvmznzxy/nemotron-3-super-120b-a12b
- OpenRouter ↗nvidia/nemotron-3-super-120b-a12b
- Model card ↗Family-level model document · NVIDIA · first party
- Release source ↗direct research preview identity date
- huggingface.co ↗
Safety evals
Rank and direction are specific to the Safety portfolio. Black marks the observed result; the grey dot marks the value implied by the Safety rank. Values stay on each source’s native scale.
| Eval / sub-eval | Rank | Value | Distribution | Source |
|---|---|---|---|---|
| AA-Omnisciencehallucination_rate | #248 / 358 | ↓0.8696 | Source ↗official | |
| Arena Factuality — Text Arena (factuality-only weighting)factuality_bt_rating | #99 / 111 | ↑1397.0 | Source ↗official | |
| BullshitBench v2clear_pushback_rate | #46 / 122 | ↑0.4375 | Source ↗official | |
| Cisco AI Defense Rolling Single-Turn Leaderboardsingle_turn_attack_success_rate | #56 / 104 | ↓40.43 | Source ↗official | |
| DystopiaBenchbasaglia_score | #29 / 50 | ↓65.33 | Source ↗official | |
| DystopiaBenchbaudrillard_score | #22 / 50 | ↓53.23 | Source ↗official | |
| DystopiaBenchhuxley_score | #23 / 50 | ↓69.47 | Source ↗official | |
| DystopiaBenchlaguardia_score | #26 / 50 | ↓67.17 | Source ↗official | |
| DystopiaBenchorwell_score | #25 / 50 | ↓68.63 | Source ↗official | |
| DystopiaBenchpetrov_score | #34 / 50 | ↓75.17 | Source ↗official | |
| Enkrypt AI Safety Leaderboardbias_attack_non_success_rate | #158 / 270 | ↑14.73 | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #257 / 270 | ↑50.83 | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #63 / 270 | ↑92.22 | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #63 / 268 | ↑98.18 | Source ↗official | |
| MT-JailBench CrescendoXsafety_score | #5 / 21 | ↑44.03 | Source ↗official | |
| Pokee-Isaac model card — DTAPbenign_task_success_rate | #6 / 6 | ↑0.633 | Source ↗official | |
| Pokee-Isaac model card — DTAPcombined_attack_success_rate | #5 / 6 | ↓0.604 | Source ↗official | |
| RefusalBenchyouden_j | #12 / 19 | ↑0.06383 | Source ↗official | |
| SM-Benchadversarial | #87 / 92 | ↑71.22 | Source ↗official | |
| SM-Benchambiguous_interpretation | #88 / 92 | ↑63.39 | Source ↗official | |
| SM-Benchanti_hallucination | #60 / 92 | ↑90.58 | Source ↗official | |
| SM-Bencheq_boundaries | #83 / 92 | ↑52.53 | Source ↗official | |
| SM-Benchoverfit | #89 / 92 | ↑21.86 | Source ↗official | |
| TACbase_welfare_rate | #35 / 92 | ↑29.5 | Source ↗self run |
Freedom evals
Rank and direction are specific to the Freedom portfolio. Black marks the observed result; the grey dot marks the value implied by the Freedom rank. Values stay on each source’s native scale.
| Eval / sub-eval | Rank | Value | Distribution | Source |
|---|---|---|---|---|
| Cisco AI Defense Rolling Single-Turn Leaderboardsingle_turn_attack_success_rate | #49 / 104 | ↑40.43 | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #13 / 270 | ↓50.83 | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #206 / 270 | ↓92.22 | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #201 / 268 | ↓98.18 | Source ↗official | |
| MT-JailBench CrescendoXsafety_score | #17 / 21 | ↓44.03 | Source ↗official | |
| SM-Benchadversarial | #6 / 92 | ↓71.22 | Source ↗official | |
| SM-Bencheq_boundaries | #83 / 92 | ↑52.53 | Source ↗official | |
| SM-Benchoverfit | #89 / 92 | ↑21.86 | Source ↗official | |
| SpeechMap model completioncomplete_pct | #136 / 181 | ↑42.5 | Source ↗official |