NVIDIAOpen weights: nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-FP8Reasoning modelCompare with another modelNemotron 3 Super 120B A12B
Nemotron 3 Super 120B A12B is a NVIDIA open-weights reasoning model, released 11 Mar 2026. Its best configuration (thinking reasoning effort) ranks #366 of 785 on the BenchLeader Index at 50.6 ±7.8, in the upper half. The ± is the point: 431 other configurations score within that range, so they and this one cannot be told apart on quality alone — price and speed are what separate them. It scores highest in instruction following (70) and lowest in knowledge (43). At $0.450 per million tokens blended it is cheaper than most ranked models. Output speed of 149 tokens per second puts it in the fastest quarter, with a first answer in 14.8 s. It has been measured at 2 reasoning-effort settings; this summary describes the best-scoring one, and the tabs above switch between them. Last measured 11 Mar 2026.stale: no new result in six months
- Blended price
- $0.450/M
- $0.300 in · $0.900 out
- Output speed
- 149 tok/s
- measured by Artificial Analysis
- First answer
- 15 s
- first token 1.03 s
- Context
- 262k
- Full answer
- 18 s
- median, reasoning included
- Cost per run
- $1.64
- one full Intelligence Index run
- Released
- 11 Mar 2026
- Overall index51
- Reasoning46
- Agents & tools45
- Knowledge43
- Instruction following70
- Long context58
- Composite44
Reasoning-effort configurations
The same model behaves differently depending on how much it is allowed to think. Each row is one setting, scored only on the benchmarks that were run at that setting. “Not stated” collects results from publishers that did not say which setting they used; for a reasoning model that is usually its thinking mode, but we do not assume it. Pick a setting here or at the top of the page to see its price, speed and category scores.
| Effort | Index | Rank | Speed | First answer | Chat reply cost | Categories |
|---|---|---|---|---|---|---|
| thinkingbest | 50.6 | #366 | 149 tok/s | 15 s | $0.0004 | Agents & tools 45 · Composite 44 · Instruction following 70 · Knowledge 43 · Long context 58 · Reasoning 46 |
| not stated | – | – | 33 tok/s | 1.03 s | $0.0008 | Coding 53 · Human preference 53 · Maths 54 · Reasoning 53 |
How its index has moved
19 Sept 2026 to 21 Oct 2026The index is recomputed from scratch every day, so a line moves when a new benchmark result lands, when a publisher revises a score, or when the models it is normalised against change. Early movement usually means the score is still settling.
Benchmark results
See all LLM benchmarks. One column per reasoning effort. Rank is among every configuration of every model on that benchmark. Hover a score for the run it came from.
Reasoning
| Benchmark | thinking | not stated | Source | Trend |
|---|---|---|---|---|
| Arena Hard Promptseffort not stated | – | 1378#189 | Arena (LMArena) | |
| GPQA Diamond (AA)not in index | 80.0%#193 | – | Artificial Analysis | |
| Humanity's Last Exam (AA)not in index | 20.8%#212 | – | Artificial Analysis | |
| CritPt | 3.1%#176 | – | Artificial Analysis |
Coding
| Benchmark | thinking | not stated | Source | Trend |
|---|---|---|---|---|
| Arena Codingeffort not stated | – | 1407#189 | Arena (LMArena) | |
| SciCode (AA)not in index | 36.2%#196 | – | Artificial Analysis | |
| Terminal-Bench 4.0 (AA)not in index | 0.0%#152 | – | Artificial Analysis | |
| Terminal-Bench 2.1 (AA)not in index | 38.6%#144 | – | Artificial Analysis |
Agents & tools
| Benchmark | thinking | Source | Trend |
|---|---|---|---|
| Terminal-Bench Hard | 28.8%#117 | Artificial Analysis | |
| τ²-Bench Telecom (AA)not in index | 67.8%#162 | Artificial Analysis | |
| τ³-Banking (AA) | 10.3%#142 | Artificial Analysis | |
| ITBench SRE (AA)not in index | 1.1%#44 | Artificial Analysis | |
| APEX-Agents (AA)not in index | 1.8%#32 | Artificial Analysis | |
| AutomationBenchnot in index | 3.8%#171 | Artificial Analysis | |
| GDP.pdf (AA)not in index | 2.6%#186 | Artificial Analysis | |
| EnterpriseOps-Gymnot in index | 23.6%#48 | Artificial Analysis |
Maths
| Benchmark | not stated | Source | Trend |
|---|---|---|---|
| Arena Maths | 1371#181 | Arena (LMArena) |
Knowledge
| Benchmark | thinking | Source | Trend |
|---|---|---|---|
| AA-Omniscience | -41.5#326 | Artificial Analysis | |
| AA-Omniscience: accuracynot in index | 24.3%#246 | Artificial Analysis | |
| AA-Omniscience: non-hallucinationnot in index | 13.0%#364 | Artificial Analysis |
Instruction following
| Benchmark | thinking | not stated | Source | Trend |
|---|---|---|---|---|
| IFBench | 71.5%#54 | – | Artificial Analysis | |
| Arena Instruction Followingnot in indexeffort not stated | – | 1342#197 | Arena (LMArena) |
Human preference
| Benchmark | not stated | Source | Trend |
|---|---|---|---|
| Arena Text | 1360#190 | Arena (LMArena) | |
| Arena Creative Writingnot in index | 1303#214 | Arena (LMArena) | |
| Arena Multi-turnnot in index | 1350#194 | Arena (LMArena) | |
| Arena Longer Queriesnot in index | 1358#198 | Arena (LMArena) |
Long context
| Benchmark | thinking | Source | Trend |
|---|---|---|---|
| AA-LCR | 65.7%#247 | Artificial Analysis |
Composite
| Benchmark | thinking | Source | Trend |
|---|---|---|---|
| AA Intelligence Index v4.3.2 | 12.8#321 | Artificial Analysis |
Where to run it
Every provider serving this model through OpenRouter, with throughput and first-token latency measured on live traffic over the last 30 minutes and each provider’s own price. Purple marks the best in each column.
| Provider | Speed | First token | Input $/M | Output $/M | Context | Quantisation |
|---|---|---|---|---|---|---|
| Crusoe | 135 tok/s | 0.27 s | $0.300 | $2.40 | 262k | fp8 |
| DeepInfra | 77 tok/s | 2.55 s | $0.085 | $0.400 | 262k | bf16 |
| DekaLLM | 30 tok/s | 1.07 s | $0.080 | $0.450 | 262k | fp8 |
Price history
The price above, per 1M tokens, every day it changed.
What a task costs
Estimates from list price, output speed and time to first answer for the best configuration. “With caching” assumes three-quarters of the input is served from the prompt cache at $0.060 per 1M. Reasoning tokens are not modelled.
| Workload | Tokens in / out | Cost | With caching | Time |
|---|---|---|---|---|
| Chat reply | 400 / 300 | $0.0004 | $0.0003 | 16.8 s |
| Summarize a 30-page report | 12,000 / 600 | $0.0041 | $0.0020 | 18.8 s |
| Code edit | 6,000 / 1,500 | $0.0032 | $0.0021 | 24.8 s |
| Agentic coding session | 60,000 / 4,000 | $0.022 | $0.011 | 41.5 s |
| Structured extraction | 2,000 / 200 | $0.0008 | $0.0004 | 16.1 s |
More from NVIDIA
See also
Questions people ask
- What are Nemotron 3 Super 120B A12B's benchmark scores?
- As of 24 Oct 2026, Nemotron 3 Super 120B A12B (thinking effort) scores 50.6 ±7.8 on the BenchLeader Index, strongest in instruction following (70) and weakest in knowledge (43). That rests on 7 independent benchmark results, and every score on this page links to its publisher.
- How does Nemotron 3 Super 120B A12B rank?
- Nemotron 3 Super 120B A12B (thinking effort) is #366 of 785 ranked configurations on 24 Oct 2026. 431 other configurations score within its margin, so the exact position matters less than the band. The leader is Claude Opus 5.5 (max effort) at 72.6.
- How much does Nemotron 3 Super 120B A12B cost per 1M tokens?
- As of 24 Oct 2026, Nemotron 3 Super 120B A12B costs $0.300 per 1M input tokens and $0.900 per 1M output tokens, $0.450 blended (three input tokens per output token). The list price last changed on 6 Oct 2026.
- When was Nemotron 3 Super 120B A12B released?
- NVIDIA released Nemotron 3 Super 120B A12B on 11 Mar 2026. BenchLeader has results for it at 2 reasoning-effort settings.
Data as of 24 Oct 2026. Compare these configurations.
Cite as: BenchLeader, “Nemotron 3 Super 120B A12B: benchmarks, pricing, speed and rank”, data as of 24 Oct 2026, https://www.benchleader.com/models/nemotron-3-super-120b-a12b.