← Back to directory
NVIDIA
Nemotron 3 Super
Released Mar 11, 2026 · knowledge cutoff unpublished
- Status
- Superseded
- Location
- United States
- Modality
- Text
- Context window
- 1M
- Max output
- —
- Speed
- —
- Price ($/MTok in / out)
- $0.10 / $0.50
- Cost per task
- $0.20
- Open weights
- Yes
Benchmarks4 of 8 reported
MMLU-Pro
83.73%
SWE-bench Verified
60.47%
AIME (latest)
90.21%
Humanity's Last Exam
18.26%
GPQA Diamond
—
Terminal-Bench 2.1
—
LMArena Elo
—
ARC-AGI-2
—
Strengths & weaknesses
Strengths
- Among the fastest open models at its capability — roughly 450 output tokens/sec on hosted endpoints
- Activates about a tenth of its parameters per token, so it serves far cheaper than its size implies
- Handles very long inputs without the slowdown dense models hit, thanks to the hybrid state-space design
Weaknesses
- Text-only, so no screenshots, diagrams or PDFs
- Reasoning depth well short of frontier flagships — a fast workhorse rather than a hard-problem model
Middle of the Nemotron 3 family; 120B total / 12.7B active MoE. MMLU-Pro 83.73, SWE-bench Verified 60.47, AIME 2025 90.21 and HLE 18.26 per llm-stats' launch analysis. GPQA Diamond left null deliberately: the reported 79.23 is labelled "GPQA no tools", which NVIDIA's own report treats as distinct from GPQA Diamond. Pricing is DeepInfra's rate; OpenRouter lists $0.085/$0.40 and the cross-provider median is about $0.30/$0.80. Artificial Analysis Intelligence Index 36.0.
For developers
API model strings
Not researched
Licence
Not researched
Retirement
No retirement announced
Lineage
No recorded predecessor