Tech Meridian← ALL MODELS
PROMY MERIDIAN

NVIDIA · MODEL RELEASE TRACKER

Nemotron 3.5 Super VL

Nemotron 3.5 Super VL is an open, customizable multimodal model from NVIDIA described as delivering frontier-level accuracy on scientific benchmarks and fast time-to-solution for scientific coding; the article describes post-training recipes that improved accuracy in a drug-metabolism example and mentions a 16,384-token output budget.

CURRENT SNAPSHOT4/5 DIMENSIONS WITH DATA

The dimensions that change the decision.

PRICING

Not established from the available sources.

CONTEXT WINDOW
Output token budget / context windowExample usage mentions a 16,384-token output budget for a drug-metabolism prediction workflow.
MODALITIES
ModalityDescribed as an open, customizable multimodal model (supports multiple input modalities).
BENCHMARKS
Single-GPU throughputOn a single GPU, delivers up to 5× higher throughput for agentic tasks compared to other open models in its size category (as reported by NVIDIA).
Post-training reported improvementsOrganizations that post-trained the model reportedly achieved improvements such as ~40× lower inference cost, ~10× lower latency, and ~40% higher accuracy for specialized workflows.
Single-GPU throughput claimOn a single GPU, Nemotron 3.5 Super VL delivers up to 5× higher throughput for agentic tasks compared to other open models in its size category (as claimed by NVIDIA).
Performance claims vs. competitorsNvidia says Super VL completes tasks faster than certain competing models (Anthropic and OpenAI models named Opus 5, GPT 6.1 Sol, Luna, Astra) when post-trained on specific agentic tasks such as document intelligence, financial accounting, scientific experimentation, medical insurance claims, and security attack resilience.
Comparison to Nemotron 3 UltraUses fewer parameters than Nemotron 3 Ultra; NVIDIA states it can deliver similar accuracy in instruction-following and tool-use tasks despite the smaller size.
Performance claim versus competitorsNVIDIA claims Super VL completes tasks faster than certain competing models (named examples include Anthropic models and OpenAI models such as Opus 5 and GPT 6.1 Sol) when post-trained on specific agentic tasks.
Reported benchmark scoresNVIDIA reports 63.44% on Hermes Bench, 87.05% on LangChain agent evaluations, and 80.8% on the CharXiv chart-reasoning benchmark.
Scientific benchmark accuracyReported to deliver 'frontier-level' accuracy on scientific benchmarks.
AVAILABILITY
Availability / opennessDescribed as an 'open' model (implying open availability or openness to customization).
Reported customers / usersNvidia listed several companies reported as using Super VL, including Skydance, EY, Ramp, Crowdstrike, Perplexity, and Salesforce.

RELEASE TIMELINE

Published, source-backed release events only.

2 SOURCES · IMPORTANCE 8.0

Nvidia unveils Nemotron 3.5 Super VL and lightweight Bolt models for enterprise agents and local AI PCs

Nvidia announced new open-weight models aimed at always-on agents and edge/PC deployment: Nemotron 3.5 Super VL (120B parameters) for enterprise agents, plus two lightweight Bolt models including Nemotron 3.5 Bolt 74B-A9B 24GB and the smaller Bolt E4B for local AI PCs and workstations. Nvidia says Super VL and Bolt 74B-A9B are available now (Bolt E4B coming later this month), claims Super VL can match larger Nemotron 3 Ultra on instruction-following and outperform some competing models in post-trained agent tasks, and named early users such as Skydance, EY, Ramp, CrowdStrike, Perplexity and Salesforce.

→
2 SOURCES · IMPORTANCE 8.0

NVIDIA releases Nemotron 3.5 Super VL, a 120B-parameter open multimodal model for agents

NVIDIA announced Nemotron 3.5 Super VL, an open 120B-parameter mixture-of-experts multimodal model that activates 12B parameters per token to support agentic workflows combining reasoning, instruction following, tool use, and text, image, document, and video understanding. NVIDIA says it achieves up to 5× higher single‑GPU throughput on agentic tasks versus comparable open models, supports post‑training via NeMo (SFT, LoRA, RL) with resources on Hugging Face and a GitHub usage cookbook, and employs inference optimizations such as LiLiCorr speculative decoding and NVFP4 quantization; partners including Scale AI, Ramp, and AI21 report substantial cost, latency, or accuracy improvements in production workloads.

→

VERIFIABLE FACTS

Every value stays attached to a source and date.

BENCHMARKS · Performance claims vs. competitorsDEVELOPER CLAIM

Nvidia says Super VL completes tasks faster than certain competing models (Anthropic and OpenAI models named Opus 5, GPT 6.1 Sol, Luna, Astra) when post-trained on specific agentic tasks such as document intelligence, financial accounting, scientific experimentation, medical insurance claims, and security attack resilience.

CAPABILITIES · Use cases and agent integrationDEVELOPER CLAIM

Targeted use cases include coding (test/validate code, analyze errors, recommend fixes), web research (cross-reference/verify sources), document intelligence (read tables, charts, page layouts), cybersecurity (identify vulnerabilities, triage threats), and video understanding (search, content review, incident analysis). Integrates with major open agent frameworks and orchestration systems via NVIDIA reference implementations and the NVIDIA Agent Toolkit.

WHAT CHANGED

Stored passport versions, without reconstructed history.

Passport updated27 facts
AVAILABILITY · Availability and deployment on AWS− Available on AWS; can be deployed on Amazon Elastic Kubernetes Service (Amazon EKS), Amazon EKS Auto Mode, and Amazon SageMaker HyperPod EKS.
CONTEXT WINDOW · Context and memory behavior− Mamba-2 layers carry a constant-size state during generation, so their memory does not grow with sequence length like KV cache does, which helps keep long contexts affordable.
MODALITIES · Modalities supported− Supports text, image, document, and video understanding; one model serves both text and visual tasks and supports agent workflows combining reasoning with visual understanding.
MODALITIES · Modality+ Described as an open, customizable multimodal model (supports multiple input modalities).
AVAILABILITY · Availability / openness+ Described as an 'open' model (implying open availability or openness to customization).
CONTEXT WINDOW · Output token budget / context window+ Example usage mentions a 16,384-token output budget for a drug-metabolism prediction workflow.
CAPABILITIES · Post-training accuracy improvements in example+ In a drug-metabolism prediction example, supervised fine-tuning improved accuracy by 12%, and reinforcement-learning post-training improved it a further 17% (as reported).
BENCHMARKS · Scientific benchmark accuracy+ Reported to deliver 'frontier-level' accuracy on scientific benchmarks.
CAPABILITIES · Time to solution for scientific coding+ Reported to provide the fastest time to solution for scientific coding (as claimed by NVIDIA).
Passport updated24 facts
CAPABILITIES · Architecture components+ Hybrid Mamba-2, MoE, and attention layers with a RADIO vision encoder; a recursively shared Multi-Token Prediction (MTP) head supports native speculative decoding.
AVAILABILITY · Availability and deployment on AWS+ Available on AWS; can be deployed on Amazon Elastic Kubernetes Service (Amazon EKS), Amazon EKS Auto Mode, and Amazon SageMaker HyperPod EKS.
RELEASE · Distillation and openness+ Open multimodal model distilled from Nemotron 3 Ultra; NVIDIA publishes the weights, training datasets, and post-training recipes.
CONTEXT WINDOW · Context and memory behavior+ Mamba-2 layers carry a constant-size state during generation, so their memory does not grow with sequence length like KV cache does, which helps keep long contexts affordable.
MODALITIES · Modalities supported− Multimodal model that handles text, images, documents, and video in a single model.
MODALITIES · Modalities supported+ Supports text, image, document, and video understanding; one model serves both text and visual tasks and supports agent workflows combining reasoning with visual understanding.
AVAILABILITY · Availability / license characterization− Described as an open-weight model intended for enterprise deployment and post-training by organizations.
CAPABILITIES · Parameters and Mixture-of-Experts behavior+ 120B-parameter Mixture-of-Experts (MoE) model; sparse routing activates 12B parameters per token.
RELEASE · Release / announcement− NVIDIA announced and released Nemotron 3.5 Super VL in October 2026.
BENCHMARKS · Reported benchmark scores+ NVIDIA reports 63.44% on Hermes Bench, 87.05% on LangChain agent evaluations, and 80.8% on the CharXiv chart-reasoning benchmark.
CAPABILITIES · Use cases and agent integration+ Targeted use cases include coding (test/validate code, analyze errors, recommend fixes), web research (cross-reference/verify sources), document intelligence (read tables, charts, page layouts), cybersecurity (identify vulnerabilities, triage threats), and video understanding (search, content review, incident analysis). Integrates with major open agent frameworks and orchestration systems via NVIDIA reference implementations and the NVIDIA Agent Toolkit.
Passport updated19 facts
BENCHMARKS · Comparison to Nemotron 3 Ultra+ Uses fewer parameters than Nemotron 3 Ultra; NVIDIA states it can deliver similar accuracy in instruction-following and tool-use tasks despite the smaller size.
CAPABILITIES · Intended useDesigned for always-on agents and edge computing; geared toward enterprise agent applications.→Positioned for use by enterprise agents (always-on agents and edge computing scenarios).
RELEASE · Launch / announcement+ Announced/launched by NVIDIA in October 2026 (reported 2026-10-19) as part of a set of new open-weight models.
AVAILABILITY · Availability / license characterization+ Described as an open-weight model intended for enterprise deployment and post-training by organizations.
CAPABILITIES · Parameter count+ Described as a large model with 120 billion parameters.
RELEASE · Release / launch date− Announced/launched on 2026-10-19 (reported by TechCrunch).
BENCHMARKS · Performance claim versus competitors+ NVIDIA claims Super VL completes tasks faster than certain competing models (named examples include Anthropic models and OpenAI models such as Opus 5 and GPT 6.1 Sol) when post-trained on specific agentic tasks.
AVAILABILITY · Weight availability− Released as an open-weight model.
Passport updated16 facts
AVAILABILITY · Reported customers / users+ Nvidia listed several companies reported as using Super VL, including Skydance, EY, Ramp, Crowdstrike, Perplexity, and Salesforce.
CAPABILITIES · Intended use / target+ Designed for always-on agents and edge computing; geared toward enterprise agent applications.
AVAILABILITY · Openness and customization− Presented as an open, customizable model; developers can post-train and customize it using NVIDIA NeMo (supervised fine-tuning, low-rank adaptation, and reinforcement learning). NVIDIA is publishing model weights, datasets, and training recipes alongside the release.
CAPABILITIES · Parameters+ 120 billion parameters.
BENCHMARKS · Performance claims vs. competitors+ Nvidia says Super VL completes tasks faster than certain competing models (Anthropic and OpenAI models named Opus 5, GPT 6.1 Sol, Luna, Astra) when post-trained on specific agentic tasks such as document intelligence, financial accounting, scientific experimentation, medical insurance claims, and security attack resilience.
RELEASE · Release / launch date+ Announced/launched on 2026-10-19 (reported by TechCrunch).
AVAILABILITY · Weight availability+ Released as an open-weight model.
Passport updated11 facts
CAPABILITIES · Agentic and reasoning capabilities+ Designed for agentic workflows: combines reasoning, instruction following, and tool use to support continuous (always-on) enterprise agents.
CAPABILITIES · Architecture and parameters+ 120-billion-parameter mixture-of-experts (MoE) architecture that activates 12 billion parameters per token.
AVAILABILITY · Customization and availability− Described as open and customizable; developers can post-train using NVIDIA NeMo with supervised fine-tuning, low-rank adaptation, and reinforcement learning on open datasets released on Hugging Face.
CAPABILITIES · Inference optimizations and techniquesUses LiLiCorr speculative decoding, NVFP4 quantization with quantization-aware distillation, multi-token prediction, and context reuse to optimize inference efficiency (including on NVIDIA DGX Spark).→Uses inference optimizations such as LiLiCorr speculative decoding, NVFP4 quantization with quantization-aware distillation, multi-token prediction, and context reuse to improve inference efficiency.
MODALITIES · Modalities supportedMultimodal: text, images, documents, and video understanding in a single model.→Multimodal model that handles text, images, documents, and video in a single model.
AVAILABILITY · Openness and customization+ Presented as an open, customizable model; developers can post-train and customize it using NVIDIA NeMo (supervised fine-tuning, low-rank adaptation, and reinforcement learning). NVIDIA is publishing model weights, datasets, and training recipes alongside the release.
RELEASE · Release / announcement+ NVIDIA announced and released Nemotron 3.5 Super VL in October 2026.
BENCHMARKS · Single-GPU throughput claim+ On a single GPU, Nemotron 3.5 Super VL delivers up to 5× higher throughput for agentic tasks compared to other open models in its size category (as claimed by NVIDIA).
Passport created7 facts