Nemotron 3.5 Lightning

Fast NVIDIA Nemotron MoE for reliable agentic tasks across enterprise workloads
- Context
- 262K
- Released
- 2mo ago
- Parameters
- 30B
- Input / 1M tokens
- $0.07
- Output / 1M tokens
- $0.20
- Cached input / 1M tokens
- $0.04
Sources:models.devOpenRouter
Benchmarks
DeepSWE is the primary coding benchmark where an exact model evaluation exists. Other sourced evaluations appear below; missing scores are not zero.
Not evaluated in DeepSWE 1.1 as of 2026-09-22. Results are only shown for exact model versions in the official dataset. Official leaderboard.
No benchmark results have been matched to this model yet.
Specifications
- Developer
- NVIDIA
- Model family
- Nemotron
- Type
- Language model
- Predecessor
- —
- Successor
- —
- Released
- August 11, 2026
- Model ID
nvidia/nemotron-3.5-lightning- Openness
- Open weights
- License
- —
- Parameters
- 30B
- Input modalities
- text
- Output modalities
- text
- Modality
- text->text
- Native context
- 1M
- Tokenizer
- Other
- Max output tokens
- 236K
- Source freshness
- Metadata synced Oct 8, 2026, 10:17 PM UTC
Additional source evidence
- Source model
- nvidia/nemotron-3.5-lightning
- Source family
- nemotron
- Last updated
- 2026-08-11
- Model context
- 262K
- Model output limit
- 262K
- Open weights
- Yes
- File attachments
- No
- Reasoning
- Yes
- Tool calling
- Yes
- Structured output
- Yes
- Temperature control
- Yes
Published benchmarks
| Benchmark | Score | Evaluation details | Evidence |
|---|---|---|---|
| MMLU-Pro | 81.94 · score | NeMo Gym / NeMo Evaluator SDK · BF16; reasoning | Source |
| AA-Omniscience | 17.5 · score | NeMo Gym / NeMo Evaluator SDK · BF16; reasoning | Source |
| SciCode | 32.6 · score | NeMo Gym / NeMo Evaluator SDK · BF16; reasoning | Source |
| PinchBench | 85.37 · score | NeMo Gym / NeMo Evaluator SDK · BF16; reasoning | Source |
| BrowseComp | 36.97 · score | NeMo Gym / NeMo Evaluator SDK · BF16; reasoning | Source |
| AA-LCR | 52 · score | NeMo Gym / NeMo Evaluator SDK · BF16; reasoning | Source |
| GPQA Diamond | 75.44 · accuracy | BF16; reasoning; no tools | Source |
| Humanity's Last Exam | 11.72 · accuracy | text-only · BF16; reasoning; no tools | Source |
| SWE-Bench Verified | 51.56 · accuracy | NeMo Evaluator · BF16; reasoning | Source |
| SWE-Bench Multilingual | 39.33 · accuracy | NeMo Evaluator · BF16; reasoning | Source |
| Terminal-Bench | 24.58 · accuracy | NeMo Evaluator · BF16; reasoning · 2.1 | Source |
| tau3-bench | 9.28 · score | BF16; reasoning; banking | Source |
| GDPval-AA | 832 · Elo | BF16; reasoning · 2 | Source |
| IFBench | 71.88 · accuracy | BF16; reasoning; loose | Source |
Complete source record (JSON)
{
"id": "nvidia/nemotron-3.5-lightning",
"name": "Nemotron 3.5 Lightning 30B A3B",
"description": "Fast NVIDIA Nemotron MoE for reliable agentic tasks across enterprise workloads",
"family": "nemotron",
"last_updated": "2026-08-11",
"attachment": false,
"reasoning": true,
"tool_call": true,
"structured_output": true,
"temperature": true,
"modalities": {
"input": [
"text"
],
"output": [
"text"
]
},
"limit": {
"context": 262144,
"output": 262144
},
"benchmarks": [
{
"name": "MMLU-Pro",
"score": 81.94,
"source": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16",
"metric": "score",
"harness": "NeMo Gym / NeMo Evaluator SDK",
"variant": "BF16; reasoning"
},
{
"name": "AA-Omniscience",
"score": 17.5,
"source": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16",
"metric": "score",
"harness": "NeMo Gym / NeMo Evaluator SDK",
"variant": "BF16; reasoning"
},
{
"name": "SciCode",
"score": 32.6,
"source": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16",
"metric": "score",
"harness": "NeMo Gym / NeMo Evaluator SDK",
"variant": "BF16; reasoning"
},
{
"name": "PinchBench",
"score": 85.37,
"source": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16",
"metric": "score",
"harness": "NeMo Gym / NeMo Evaluator SDK",
"variant": "BF16; reasoning"
},
{
"name": "BrowseComp",
"score": 36.97,
"source": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16",
"metric": "score",
"harness": "NeMo Gym / NeMo Evaluator SDK",
"variant": "BF16; reasoning"
},
{
"name": "AA-LCR",
"score": 52,
"source": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16",
"metric": "score",
"harness": "NeMo Gym / NeMo Evaluator SDK",
"variant": "BF16; reasoning"
},
{
"name": "GPQA Diamond",
"score": 75.44,
"source": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16",
"metric": "accuracy",
"variant": "BF16; reasoning; no tools"
},
{
"name": "Humanity's Last Exam",
"score": 11.72,
"source": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16",
"metric": "accuracy",
"variant": "BF16; reasoning; no tools",
"dataset": "text-only"
},
{
"name": "SWE-Bench Verified",
"score": 51.56,
"source": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16",
"metric": "accuracy",
"harness": "NeMo Evaluator",
"variant": "BF16; reasoning"
},
{
"name": "SWE-Bench Multilingual",
"score": 39.33,
"source": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16",
"metric": "accuracy",
"harness": "NeMo Evaluator",
"variant": "BF16; reasoning"
},
{
"name": "Terminal-Bench",
"score": 24.58,
"source": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16",
"metric": "accuracy",
"harness": "NeMo Evaluator",
"variant": "BF16; reasoning",
"version": "2.1"
},
{
"name": "tau3-bench",
"score": 9.28,
"source": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16",
"metric": "score",
"variant": "BF16; reasoning; banking"
},
{
"name": "GDPval-AA",
"score": 832,
"source": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16",
"metric": "Elo",
"variant": "BF16; reasoning",
"version": "2"
},
{
"name": "IFBench",
"score": 71.88,
"source": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16",
"metric": "accuracy",
"variant": "BF16; reasoning; loose"
}
],
"release_date": "2026-08-11",
"open_weights": true
}Infrastructure
Provider availability
| Provider | Endpoint | Context | Input /M tokens | Output /M tokens | Cached input /M tokens | Cache write /M tokens | Uptime 1d | Data handling |
|---|---|---|---|---|---|---|---|---|
| darkbloom/int4 | 262K | $0.02 | $0.12 | $0.01 | — | 99.07% | ZDR: NoTraining: Unknown Prompt logging: Unknown Retention: Unknown Checked 2026-10-15 | |
| wafer | 262K | $0.04 | $0.13 | $0.02 | — | 100.00% | ZDR: YesTraining: Unknown Prompt logging: Unknown Retention: Unknown Checked 2026-10-15 | |
| io-net | 262K | $0.05 | $0.13 | $0.02 | — | 100.00% | ZDR: YesTraining: Unknown Prompt logging: Unknown Retention: Unknown Checked 2026-10-15 | |
| deepinfra/bf16 | 262K | $0.06 | $0.16 | $0.03 | — | 100.00% | ZDR: YesTraining: Unknown Prompt logging: Unknown Retention: Unknown Checked 2026-10-15 | |
| coreweave/bf16 | 262K | $0.07 | $0.20 | $0.04 | — | 100.00% | ZDR: YesTraining: Unknown Prompt logging: Unknown Retention: Unknown Checked 2026-10-15 |
Provider data:OpenRouter
Deployment graph
Can I run this?
Vertical Tabs only shows compatibility when it has structured evidence tying this exact model to hardware, runtime, quantization, memory use, and observed performance.