N
Nemotron 3 Super
Open SourceNVIDIA•Released on 2026-03-13
NVIDIA's flagship open-source model for agentic AI, featuring 120B total parameters with 12B active (MoE). Hybrid Mamba-Transformer architecture delivers 5x throughput vs previous Nemotron Super. 1M context window prevents goal drift in complex multi-agent workflows. #1 on DeepResearch Bench.
82
Overall Score
Core Specs
1000K
Context Window
40K
Max Output
✓
Reasoning
✓
Open Source
Multimodal Support
text
Scenario Scores
User Feedback Highlights
Based on community feedback. Hover to see original reviews.
+ Multi-token prediction for 3x faster inference+ Open weights under permissive license+ #1 on DeepResearch Bench I & II− New release, limited community feedback− Text-only (no multimodal support)− Requires high-end hardware for self-hosting+ 1M context window for full workflow state+ 5x throughput vs previous Nemotron Super
Sentiment:👍 100%😐 0%👎 0%
Pros & Cons
Pros
- +1M context window for full workflow state
- +5x throughput vs previous Nemotron Super
- +Open weights under permissive license
- +#1 on DeepResearch Bench I & II
- +Multi-token prediction for 3x faster inference
Cons
- −Text-only (no multimodal support)
- −Requires high-end hardware for self-hosting
- −New release, limited community feedback
Reliability
Incidents (30d)0
Open-source, reliability depends on hosting provider
Pricing
Input (per 1M tokens)$0.40
Output (per 1M tokens)$2.20
Free trial available
Updated on 2026-03-13
Get Started
1Visit the provider's website
2Create an account
3Start using the model
Compare with Others
Benchmarks
deepResearchBench1%
artificialAnalysisEfficiency1%