Language modelOpen weights

Nemotron 3.5 Super VL 120B A12B

NVIDIA

Released: October 15, 2026
Data date: October 24, 2026

Benchmark-Meta-Score: Unavailable
No independent results
Price: $0.15 · $0.5
Input · Output / 1M tokens
Input: Text, images, and video
Output: Text

Nemotron 3.5 Super VL is a multimodal MoE model with 120 billion total parameters and 12 billion active parameters. NVIDIA combines Mamba-2, Transformer layers, 512 routed experts, and a vision encoder. It accepts text, images, and video and returns text.

The Hugging Face config specifies exactly 1,048,576 context tokens. NVIDIA publishes BF16 and NVFP4 weights under OpenMDW 1.1. OpenRouter also lists API offers with provider-specific prices, context and output limits. Benchmarks for the weight variants and distinct NIM evaluations remain separate.

Specifications and access

Specification	Value and source
Provider model ID	nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-BF16Source
Access	Open weights; local deployment and NVIDIA NIMSource
Input modalities	Text, images, and videoSource
Output modalities	TextSource
License	OpenMDW 1.1Source
Context window	The model card says 1M tokens. The published BF16 config sets max_position_embeddings to exactly 1,048,576; specific serving recipes may use a shorter limit for memory reasons.Source
Architecture	Hybrid Mamba-2, Transformer, and MoE architecture with 512 routed experts, a C-RADIO vision encoder, and an MTP head.Source
Video input	MP4 video up to 2 minutes. The model-card example uses up to 256 frames at 2 FPS or 128 frames at 1 FPS for 1080p. These sampling values belong to the serving recipe; no separate universal frame limit is stated.Source
Reasoning and tools	Reasoning is controlled through enable_thinking. Tool calling and agentic workflows are documented.Source
Data cutoff	The model card lists May 2026 as the post-training data cutoff.Source
Quantized release	An NVFP4 variant is also published. Its benchmark results are recorded separately from BF16 in the research data.Source
NVIDIA NIM hosted price	No ongoing API price is documented for NVIDIA NIM. Open weights do not imply free hosted API access.Source
OpenRouter API offer	Together lists nvidia/nemotron-3.5-super-vl through OpenRouter at USD 0.15 input, USD 0.50 output and USD 0.02 cache read per million tokens. This endpoint lists 262,144 context tokens and up to 235,929 output tokens. Other providers and their unmodified status codes are listed separately.Source
Parameter count	120B total; 12B activeSource
Context window in tokens	1,048,576Source
License	OpenMDW-1.1Source
Architecture	Hybrid Mamba-2/Transformer MoE; 512 routed experts; C-RADIO vision encoder; native MTPSource
Reasoning mode	Configurable enable_thinking; regular, medium-effort, and reasoning-off modesSource
Hosted service price	NVIDIA NIM hosted price not established; OpenRouter provider tariffs documented separatelySource
Release	2026-10-15Source
Source access	Open weights, locally deployable; downloadable NVIDIA NIM; no continuing hosted API price establishedSource
Input	Text, images, videoSource
Output modality	TextSource
Reasoning control	Configurable reasoning via enable_thinking; tool calling supportedSource
Post-training data cutoff	2026-05Source
Video input limits	MP4 video up to 2 minutes; model-card sampling examples use up to 256 frames at 2 FPS or 128 frames at 1 FPS for 1080p (The frame rate and count are serving-recipe examples, not a separately stated universal frame limit.)Source
Unlabeled AA-Briefcase-Lite components	BF16: 1137 (-) \| 1252 (-) \| 40; NVFP4: 1137 (-) \| 1188 (-) \| 45 (Retained only as source text. Labels, units, and meanings of the three components are unknown, so these values are not benchmark rows.)Source
Provider model ID	OpenRouter model ID: nvidia/nemotron-3.5-super-vl OpenRouter model ID: nvidia/nemotron-3.5-super-vl:free Hugging Face bridge ID listed by OpenRouter catalog: nvidia/NVIDIA-Nemotron-3.5-Super-VLSource
Deployment options	Darkbloom [darkbloom/nvfp4]; endpoint name=Darkbloom \| nvidia/nemotron-3.5-super-vl-20261015; quantization=nvfp4; raw status code=-2 Crusoe [crusoe/bf16]; endpoint name=Crusoe \| nvidia/nemotron-3.5-super-vl-20261015; quantization=bf16; raw status code=0 Io Net [io-net/nvfp4]; endpoint name=Io Net \| nvidia/nemotron-3.5-super-vl-20261015; quantization=nvfp4; raw status code=0 Together [together]; endpoint name=Together \| nvidia/nemotron-3.5-super-vl-20261015; quantization=unknown; raw status code=0 DeepInfra [deepinfra/fp4]; endpoint name=DeepInfra \| nvidia/nemotron-3.5-super-vl-20261015; quantization=fp4; raw status code=0 CoreWeave [coreweave/nvfp4]; endpoint name=CoreWeave \| nvidia/nemotron-3.5-super-vl-20261015; quantization=nvfp4; raw status code=0 BaseTen [baseten/fp4]; endpoint name=BaseTen \| nvidia/nemotron-3.5-super-vl-20261015; quantization=fp4; raw status code=0 Venice [venice/fp4]; endpoint name=Venice \| nvidia/nemotron-3.5-super-vl-20261015; quantization=fp4; raw status code=0 Nvidia [nvidia]; endpoint name=Nvidia \| nvidia/nemotron-3.5-super-vl-20261015:free; quantization=unknown; raw status code=-2 (This lists every provider row for the exact OpenRouter route(s) queried. Status codes are preserved as returned, not interpreted or used to filter offers.)Source
Context values by source	Darkbloom [darkbloom/nvfp4]; context_length=262144 tokens Crusoe [crusoe/bf16]; context_length=262144 tokens Io Net [io-net/nvfp4]; context_length=262144 tokens Together [together]; context_length=262144 tokens DeepInfra [deepinfra/fp4]; context_length=262144 tokens CoreWeave [coreweave/nvfp4]; context_length=1048576 tokens BaseTen [baseten/fp4]; context_length=1048576 tokens Venice [venice/fp4]; context_length=262144 tokens Nvidia [nvidia]; context_length=1000000 tokens (These are limits per OpenRouter provider route. They do not replace separately documented direct-API or model-configuration values.)Source
Source output limit	Darkbloom [darkbloom/nvfp4]; max_completion_tokens=32768 tokens Crusoe [crusoe/bf16]; max_completion_tokens=235929 tokens Io Net [io-net/nvfp4]; max_completion_tokens=131072 tokens Together [together]; max_completion_tokens=235929 tokens DeepInfra [deepinfra/fp4]; max_completion_tokens=235929 tokens CoreWeave [coreweave/nvfp4]; max_completion_tokens=943718 tokens BaseTen [baseten/fp4]; max_completion_tokens=943718 tokens Venice [venice/fp4]; max_completion_tokens=32768 tokens Nvidia [nvidia]; max_completion_tokens=131072 tokens (These are output limits for the corresponding OpenRouter provider route, not direct-API limits. Provider differences remain separate.)Source
API pricing in USD per million tokens	Darkbloom [darkbloom/nvfp4]; prompt USD/1M=0.06; completion USD/1M=0.35; cache-read USD/1M=0.01; raw discount field=0 Crusoe [crusoe/bf16]; prompt USD/1M=0.1; completion USD/1M=0.4; cache-read USD/1M=0.08; raw discount field=0 Io Net [io-net/nvfp4]; prompt USD/1M=0.15; completion USD/1M=0.5; cache-read USD/1M=0.02; raw discount field=0 Together [together]; prompt USD/1M=0.15; completion USD/1M=0.5; cache-read USD/1M=0.02; raw discount field=0 DeepInfra [deepinfra/fp4]; prompt USD/1M=0.15; completion USD/1M=0.5; cache-read USD/1M=0.02; raw discount field=0 CoreWeave [coreweave/nvfp4]; prompt USD/1M=0.15; completion USD/1M=0.5; cache-read USD/1M=0.02; raw discount field=0 BaseTen [baseten/fp4]; prompt USD/1M=0.2; completion USD/1M=0.5; cache-read USD/1M=0.04; raw discount field=0 Venice [venice/fp4]; prompt USD/1M=0.25; completion USD/1M=1.25; cache-read USD/1M=0.125; raw discount field=0 Nvidia [nvidia]; prompt USD/1M=0; completion USD/1M=0; cache-read USD/1M=not reported; raw discount field=0 (Per-token API prices were multiplied exactly by 1,000,000. Raw discount fields and returned pricing schedules are preserved separately. No endpoint pricing object in these returned rows contained an input_cache_write/cache-write price; it is not reported as zero.)Source
API capabilities	Darkbloom [darkbloom/nvfp4]; supported_parameters=["reasoning","include_reasoning","temperature","top_p","top_k","frequency_penalty","presence_penalty","repetition_penalty","stop","seed","max_tokens","response_format","structured_outputs","tools","tool_choice"]; supports_tool_choice={"auto":true,"function":true,"none":true,"required":true}; supports_implicit_caching=true; native_tools={}; supports_voice_cloning=false; supports_multiple_audio_references=false; supports_image_reference=false; supports_diarization=false Crusoe [crusoe/bf16]; supported_parameters=["reasoning","include_reasoning","max_tokens","temperature","top_p","stop","frequency_penalty","presence_penalty","seed","min_p","repetition_penalty","top_k","logit_bias","response_format","structured_outputs","tools","tool_choice"]; supports_tool_choice={"auto":true,"function":false,"none":true,"required":false}; supports_implicit_caching=true; native_tools={}; supports_voice_cloning=false; supports_multiple_audio_references=false; supports_image_reference=false; supports_diarization=false Io Net [io-net/nvfp4]; supported_parameters=["reasoning","include_reasoning","max_tokens","temperature","tools","tool_choice","logprobs","top_logprobs"]; supports_tool_choice={"auto":true,"function":false,"none":true,"required":false}; supports_implicit_caching=true; native_tools={}; supports_voice_cloning=false; supports_multiple_audio_references=false; supports_image_reference=false; supports_diarization=false Together [together]; supported_parameters=["reasoning","include_reasoning","max_tokens","temperature","top_p","stop","frequency_penalty","presence_penalty","top_k","repetition_penalty","logit_bias","min_p","response_format","structured_outputs","tools","tool_choice","logprobs","top_logprobs"]; supports_tool_choice={"auto":true,"function":true,"none":true,"required":true}; supports_implicit_caching=true; native_tools={}; supports_voice_cloning=false; supports_multiple_audio_references=false; supports_image_reference=false; supports_diarization=false DeepInfra [deepinfra/fp4]; supported_parameters=["reasoning","include_reasoning","max_tokens","temperature","top_p","stop","frequency_penalty","presence_penalty","repetition_penalty","top_k","seed","min_p","logit_bias","response_format","structured_outputs","tools","tool_choice"]; supports_tool_choice={"auto":true,"function":false,"none":true,"required":false}; supports_implicit_caching=true; native_tools={}; supports_voice_cloning=false; supports_multiple_audio_references=false; supports_image_reference=false; supports_diarization=false CoreWeave [coreweave/nvfp4]; supported_parameters=["reasoning","include_reasoning","max_tokens","temperature","top_p","top_k","repetition_penalty","frequency_penalty","presence_penalty","stop","seed","response_format","structured_outputs","tools","tool_choice","logprobs","top_logprobs"]; supports_tool_choice={"auto":true,"function":true,"none":true,"required":true}; supports_implicit_caching=true; native_tools={}; supports_voice_cloning=false; supports_multiple_audio_references=false; supports_image_reference=false; supports_diarization=false BaseTen [baseten/fp4]; supported_parameters=["reasoning","include_reasoning","max_tokens","temperature","top_p","stop","frequency_penalty","presence_penalty","tools","tool_choice"]; supports_tool_choice={"auto":true,"function":true,"none":true,"required":true}; supports_implicit_caching=true; native_tools={}; supports_voice_cloning=false; supports_multiple_audio_references=false; supports_image_reference=false; supports_diarization=false Venice [venice/fp4]; supported_parameters=["reasoning","include_reasoning","max_tokens","temperature","top_p","stop","frequency_penalty","presence_penalty","top_k","response_format","structured_outputs","tools","tool_choice"]; supports_tool_choice={"auto":true,"function":false,"none":true,"required":false}; supports_implicit_caching=true; native_tools={}; supports_voice_cloning=false; supports_multiple_audio_references=false; supports_image_reference=false; supports_diarization=false Nvidia [nvidia]; supported_parameters=["reasoning","include_reasoning","temperature","max_tokens","seed","top_p","tools","tool_choice","response_format","structured_outputs"]; supports_tool_choice={"auto":true,"function":true,"none":true,"required":true}; supports_implicit_caching=false; native_tools={}; supports_voice_cloning=false; supports_multiple_audio_references=false; supports_image_reference=false; supports_diarization=false (Supported parameters and returned API capability flags are listed per provider route. The route-level architecture returned in the same response is: {"input_modalities":["text","image","video"],"instruct_type":null,"modality":"text+image+video->text","output_modalities":["text"],"tokenizer":"Other"}.)Source
Quantized variant	Darkbloom [darkbloom/nvfp4]; tag=darkbloom/nvfp4; quantization=nvfp4 Crusoe [crusoe/bf16]; tag=crusoe/bf16; quantization=bf16 Io Net [io-net/nvfp4]; tag=io-net/nvfp4; quantization=nvfp4 Together [together]; tag=together; quantization=unknown DeepInfra [deepinfra/fp4]; tag=deepinfra/fp4; quantization=fp4 CoreWeave [coreweave/nvfp4]; tag=coreweave/nvfp4; quantization=nvfp4 BaseTen [baseten/fp4]; tag=baseten/fp4; quantization=fp4 Venice [venice/fp4]; tag=venice/fp4; quantization=fp4 Nvidia [nvidia]; tag=nvidia; quantization=unknown (Tags and quantization values are taken literally from endpoint responses. Missing values are not inferred from names.)Source

Benchmark meta-analysis

Overall and area scores

These Gradually indexes place selected published measurements on a 0 to 100 scale. They are not accuracy percentages and do not guarantee results in your application.

Overall index

Not enough independent results for a score yet

provisional

Test series: 0
Test families: 0 / 65
Sources: 0
Share of the fixed test panel (informative): 0 %

Documented method · Included test families and measurements

Method version: benchmark-meta-score-v3

The index includes active measurements from reviewed primary sources when at least five models are compared in the same documented comparison slice across dataset, harness, agent, tools, and trial count. Each benchmark series and evaluator contributes one shared setup slice from the newest capture. When one source and series include several explicitly documented reasoning tiers, the highest documented tier (none through max) is selected, never the best result value. The explicit source label “Non-reasoning” is treated as none. Unknown or ambiguous tiers exclude the model from that series. Within each series, percentages and rates are logit-transformed, other units stay unchanged, and all values are then standardized. A one-factor model then estimates one shared strength per model and, per series, how sharply it separates the models. Area scores move away from that strength only as far as the model's own results in the area support. The overall score weights the core areas at 75% and finance, law and health at 25%. It requires at least eight test series from two sources and three areas. The 0 to 100 scale refers to a fixed reference group of 29 language models whose average sits at 50.

Published model-specific parameters can differ. This is not a controlled experiment; comparisons with such differences are marked provisional.

Unmeasured series are not entered as zero. Where a model has no own results in a core area, its overall strength stands in, and the note “estimated without own results in …” names the area. Fewer test series lead to a larger standard error.

The 95% interval follows from the standard error of 200 recalculations with randomly redrawn test families. It describes how precisely the recorded test series determine the value. It is no prediction for future tasks or for later recalculations with other participants.

The source count is not a count of independent experiments.

OECD handbook on composite indicators

All 0 test families and 0 of 0 raw measurements are listed here. The complete calculation for every model is available in the JSON export.

Benchmark sources at a glance

Each evaluator measures different capabilities under its own conditions. Test series, composite indices and their components stay separate. The detailed results below retain their scales, configurations and source dates.

Evaluator	Test series	Composite indices	Recorded active measurements
Artificial Analysis	13	7	189
LMArena	0	0	35

Coverage is not a quality ranking. Overlapping tests and multiple settings do not count as independent experiments. Vendor claims, historical and disputed results are excluded from this overview and remain labelled in the detailed evidence.

199 measurements from these sources still lack a verified test-family identity and do not increase the test-series count.

Measurements without matching peer values

These measurements have no matching peer values under the same test conditions. Their original values and sources remain available here.

SWE-bench Verified (Show measurement, test conditions, and source)

Source value: 71.1
Score: 71.1
Metric: percent
Unit: %
Category: Agentic
Direction: Higher is better
Source type: vendor-reported
Evaluator: NVIDIA Nemotron 3.5 Super and Lightning technical report
Status: active
Retrieved at: 2026-10-24
Methodology: Vendor-reported result. Table 10; Nemotron 3.5 Super 120B-A12B column; open-model comparison suite; default evaluation settings from linked NeMo Gym recipes.
Context: Table 10 values are a separate evaluation setup from the later released-NVFP4 and NIM comparisons.

Source NVIDIA Nemotron 3.5 Super and Lightning technical report

SWE-bench Multilingual (Show measurement, test conditions, and source)

Source value: 67.72
Score: 67.72
Metric: percent
Unit: %
Category: Agentic
Direction: Higher is better
Source type: vendor-reported
Evaluator: NVIDIA Nemotron 3.5 Super and Lightning technical report
Status: active
Retrieved at: 2026-10-24
Methodology: Vendor-reported result. Table 10; Nemotron 3.5 Super 120B-A12B column; open-model comparison suite; default evaluation settings from linked NeMo Gym recipes.
Context: Table 10 values are a separate evaluation setup from the later released-NVFP4 and NIM comparisons.

Source NVIDIA Nemotron 3.5 Super and Lightning technical report

SWE-bench Pro (Show measurement, test conditions, and source)

Source value: 51.78
Score: 51.78
Metric: percent
Unit: %
Category: Agentic
Direction: Higher is better
Source type: vendor-reported
Evaluator: NVIDIA Nemotron 3.5 Super and Lightning technical report
Status: active
Retrieved at: 2026-10-24
Methodology: Vendor-reported result. Table 10; Nemotron 3.5 Super 120B-A12B column; open-model comparison suite; default evaluation settings from linked NeMo Gym recipes.
Context: Table 10 values are a separate evaluation setup from the later released-NVFP4 and NIM comparisons.

Source NVIDIA Nemotron 3.5 Super and Lightning technical report

Tau2 Retail (Show measurement, test conditions, and source)

Source value: 69.24
Score: 69.24
Metric: percent
Unit: %
Category: Agentic
Direction: Higher is better
Source type: vendor-reported
Evaluator: NVIDIA Nemotron 3.5 Super and Lightning technical report
Status: active
Retrieved at: 2026-10-24
Methodology: Vendor-reported result. Table 10; Nemotron 3.5 Super 120B-A12B column; open-model comparison suite; default evaluation settings from linked NeMo Gym recipes.
Context: Table 10 values are a separate evaluation setup from the later released-NVFP4 and NIM comparisons.

Source NVIDIA Nemotron 3.5 Super and Lightning technical report

Tau2 Airline (Show measurement, test conditions, and source)

Source value: 76.5
Score: 76.5
Metric: percent
Unit: %
Category: Agentic
Direction: Higher is better
Source type: vendor-reported
Evaluator: NVIDIA Nemotron 3.5 Super and Lightning technical report
Status: active
Retrieved at: 2026-10-24
Methodology: Vendor-reported result. Table 10; Nemotron 3.5 Super 120B-A12B column; open-model comparison suite; default evaluation settings from linked NeMo Gym recipes.
Context: Table 10 values are a separate evaluation setup from the later released-NVFP4 and NIM comparisons.

Source NVIDIA Nemotron 3.5 Super and Lightning technical report

Tau2 Telecom (Show measurement, test conditions, and source)

Source value: 73.19
Score: 73.19
Metric: percent
Unit: %
Category: Agentic
Direction: Higher is better
Source type: vendor-reported
Evaluator: NVIDIA Nemotron 3.5 Super and Lightning technical report
Status: active
Retrieved at: 2026-10-24
Methodology: Vendor-reported result. Table 10; Nemotron 3.5 Super 120B-A12B column; open-model comparison suite; default evaluation settings from linked NeMo Gym recipes.
Context: Table 10 values are a separate evaluation setup from the later released-NVFP4 and NIM comparisons.

Source NVIDIA Nemotron 3.5 Super and Lightning technical report

Tau3 Banking (Show measurement, test conditions, and source)

Source value: 18.88
Score: 18.88
Metric: percent
Unit: %
Category: Agentic
Direction: Higher is better
Source type: vendor-reported
Evaluator: NVIDIA Nemotron 3.5 Super and Lightning technical report
Status: active
Retrieved at: 2026-10-24
Methodology: Vendor-reported result. Table 10; Nemotron 3.5 Super 120B-A12B column; open-model comparison suite; default evaluation settings from linked NeMo Gym recipes.
Context: Table 10 values are a separate evaluation setup from the later released-NVFP4 and NIM comparisons.

Source NVIDIA Nemotron 3.5 Super and Lightning technical report

Terminal-Bench 2.1 (Show measurement, test conditions, and source)

Source value: 60.67
Score: 60.67
Metric: percent
Unit: %
Category: Agentic
Direction: Higher is better
Source type: vendor-reported
Evaluator: NVIDIA Nemotron 3.5 Super and Lightning technical report
Status: active
Retrieved at: 2026-10-24
Methodology: Vendor-reported result. Table 10; Nemotron 3.5 Super 120B-A12B column; open-model comparison suite; default evaluation settings from linked NeMo Gym recipes.
Context: Table 10 values are a separate evaluation setup from the later released-NVFP4 and NIM comparisons.

Source NVIDIA Nemotron 3.5 Super and Lightning technical report

BrowseComp (Show measurement, test conditions, and source)

Source value: 78.91
Score: 78.91
Metric: percent
Unit: %
Category: Agentic
Direction: Higher is better
Source type: vendor-reported
Evaluator: NVIDIA Nemotron 3.5 Super and Lightning technical report
Status: active
Retrieved at: 2026-10-24
Methodology: Vendor-reported result. Table 10; Nemotron 3.5 Super 120B-A12B column; open-model comparison suite; default evaluation settings from linked NeMo Gym recipes.
Context: Table 10 values are a separate evaluation setup from the later released-NVFP4 and NIM comparisons.

Source NVIDIA Nemotron 3.5 Super and Lightning technical report

BrowseComp OpenCode (Show measurement, test conditions, and source)

Source value: 72.35
Score: 72.35
Metric: percent
Unit: %
Category: Agentic
Direction: Higher is better
Source type: vendor-reported
Evaluator: NVIDIA Nemotron 3.5 Super and Lightning technical report
Status: active
Retrieved at: 2026-10-24
Methodology: Vendor-reported result. Table 10; Nemotron 3.5 Super 120B-A12B column; open-model comparison suite; default evaluation settings from linked NeMo Gym recipes.
Context: Table 10 values are a separate evaluation setup from the later released-NVFP4 and NIM comparisons.

Source NVIDIA Nemotron 3.5 Super and Lightning technical report

SciCode subtask (Show measurement, test conditions, and source)

Source value: 47.13
Score: 47.13
Metric: percent
Unit: %
Category: Code
Direction: Higher is better
Source type: vendor-reported
Evaluator: NVIDIA Nemotron 3.5 Super and Lightning technical report
Status: active
Retrieved at: 2026-10-24
Methodology: Vendor-reported result. Table 10; Nemotron 3.5 Super 120B-A12B column; open-model comparison suite; default evaluation settings from linked NeMo Gym recipes.
Context: Table 10 values are a separate evaluation setup from the later released-NVFP4 and NIM comparisons.

Source NVIDIA Nemotron 3.5 Super and Lightning technical report

SciCode problem (Show measurement, test conditions, and source)

Source value: 15.1
Score: 15.1
Metric: percent
Unit: %
Category: Code
Direction: Higher is better
Source type: vendor-reported
Evaluator: NVIDIA Nemotron 3.5 Super and Lightning technical report
Status: active
Retrieved at: 2026-10-24
Methodology: Vendor-reported result. Table 10; Nemotron 3.5 Super 120B-A12B column; open-model comparison suite; default evaluation settings from linked NeMo Gym recipes.
Context: Table 10 values are a separate evaluation setup from the later released-NVFP4 and NIM comparisons.

Source NVIDIA Nemotron 3.5 Super and Lightning technical report

Page 1 of 97

Sources and data date

Type	Evidence and data date
Price	Base API prices without caching or batch discounts. Higher context tiers and other rates are listed under Costs. Source: Pricing · OpenRouter Nemotron 3.5 Super VL provider endpoints (retrieved October 24, 2026) · OpenRouter Nemotron 3.5 Super VL provider endpoints · Editorial description reviewed October 24, 2026
Input	Documented modalities. Retrieved October 24, 2026. Source: NVIDIA Nemotron 3.5 Super VL BF16 model card · NVIDIA Nemotron 3.5 Super VL BF16 model card (retrieved October 24, 2026) · NVIDIA Nemotron 3.5 Super VL BF16 model card · Editorial description reviewed October 24, 2026
Output	Documented modalities. Retrieved October 24, 2026. Source: NVIDIA Nemotron 3.5 Super VL BF16 model card · NVIDIA Nemotron 3.5 Super VL BF16 model card (retrieved October 24, 2026) · NVIDIA Nemotron 3.5 Super VL BF16 model card · Editorial description reviewed October 24, 2026
Research status	Research date October 24, 2026. 158 source URLs checked. This documents the inspected sources, not an exhaustive inventory of every publication.
Unresolved	Differences in the NIM benchmark results Note NIM model-card scores sometimes differ from the technical report (e.g. Terminal-Bench 2.1: 66.70/60.80 in NIM vs 60.67/62.78 in report Table 23). They are preserved as separate harness/config records; public materials do not fully identify the cause of each difference.
Unresolved	AA-Briefcase-Lite component labels Note The model-card row exposes three values per variant but no component labels, units, or direction in the captured table. The six unclassified numeric cells are retained as a source fact, not assigned to guessed benchmark components.
Not published	NVIDIA NIM hosted API pricing Note NVIDIA NIM documentation does not establish a continuing hosted price. Separate OpenRouter provider tariffs are documented and do not establish a NVIDIA NIM rate.
Not published	Maximum output length Note No model-wide maximum generated-token count was found in the reviewed BF16/NVFP4 cards, NIM card, or technical report. Task-specific evaluation limits are not a model-wide output cap. OpenRouter provider-specific output limits are documented separately and do not establish a model-wide ceiling.
Unresolved	GDPval-AA v2 comparator score unit Note Table 7 does not specify the scale for 34.48. The report elsewhere describes GDPval-AA v2 as an index; the numeric unit is therefore left unknown.
Not published	Individual benchmark values in Figures 5 and 10 Note The report does not print exact values for the individual bars. Figure 5 gives an aggregate nine-benchmark mean for the Super VL BF16 parent, but no per-benchmark values; these are not reconstructed from chart pixels.
Unresolved	Current OpenCompass results Note The current public OpenCompass results could not be fully retrieved during the source check. Possible results for this exact model therefore remain unresolved.
Not found in the inspected sources	Exact-model ARC-AGI-3 results Note The official current ARC-AGI-3 v3 dataset and result tables were read; no exact model group or evaluation row for this target appears in v3. This does not remove its v1/v2 scores.
Not found in the inspected sources	Exact-model Scale Labs results Note The Scale Labs index and current public result boards in the bounded capture set were read; no exact row for this target was found. Kimi-K3 and undated DeepSeek V4.1 Flash rows are not bridged to newer targets.
Additional source	NVIDIA Nemotron 3.5 Super VL NVFP4 model card (retrieved October 24, 2026)
Additional source	NVIDIA NIM model card (retrieved October 24, 2026)
Additional source	NVIDIA Nemotron 3.5 Super and Lightning technical report (retrieved October 24, 2026) · NVIDIA Nemotron 3.5 Super and Lightning technical report · Editorial description reviewed October 24, 2026
Additional source	NVIDIA Nemotron 3.5 Super and Lightning release page (retrieved October 24, 2026)
Additional source	Artificial Analysis (retrieved October 24, 2026)
Additional source	LiveBench (retrieved October 24, 2026)
Additional source	LiveBench (retrieved October 24, 2026)
Additional source	LiveBench (retrieved October 24, 2026)
Additional source	LMArena (retrieved October 24, 2026)
Additional source	LMArena (retrieved October 24, 2026)
Additional source	LMArena (retrieved October 24, 2026)
Additional source	NVIDIA Nemotron 3.5 Bolt 74B-A9B 24GB technical report (retrieved October 24, 2026)
Additional source	ARC Prize official leaderboard (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)

Specifications and access

Overall and area scores

Overall index

Benchmark sources at a glance

Measurements without matching peer values

Published prices

Sources and data date