VERIFIED ORGANISATION

NVIDIA

Explore the documented models, published evaluations and primary sources associated with this organisation.

01 / MODELS

Verified catalogue

NVIDIA Nemotron 3 Ultra

Open 550B-parameter MoE model for orchestration, research, code, and long-duration agents.

18 SEP 2026 COMPACT LOCAL ASSISTANTS

Compact NVIDIA model released on October 23, 2026, post-trained from Gemma-4-E4B-it. It has approximately 8 billion total and 4 billion effective parameters; E4B does not mean an MoE with four billion active parameters. We group BF16 and GGUF Q4_K_M in one profile, with results separated by precision. The documented modality is text to text: targeting voice assistants does not make it an audio recognition or synthesis system.

24 OCT 2026 VISUAL RENDERING OF SIMULATORS

Cosmos3-Nano-Sim-Transfer

NVIDIA distilled causal model published on October 20, 2026, to convert visual simulation controls into RGB video. Geometry, physics, and ground truth still belong to the classic simulator; the network provides visual appearance. It accepts text and depth, segmentation, edge, or blur controls. It is not a policy that produces robot actions nor a certified physics engine.

21 OCT 2026 ACTION-CONDITIONED SIMULATION

Cosmos3-Nano-Sim-Bimanual

NVIDIA causal model for visually simulating bimanual manipulation and camera navigation, published on October 20, 2026. It receives an initial image, robot actions, camera trajectory, and optional text. Its output is video, not actions: it should not be confused with a control policy. The checkpoint declares an AgiBot body; other morphologies require specific validation.

21 OCT 2026 MULTIVIEW GENERATION FOR DRIVING

Cosmos3-Nano-Transfer-Auto

NVIDIA bidirectional model for generating driving scenes with synchronized multiview video and LiDAR. Published on October 20, 2026, it combines text, WSM world controls, and an optional initial image. It supports up to eleven standard or fisheye cameras and up to 300 frames per view. It is batch generation, not the causal simulation model Sim-Auto.

21 OCT 2026 CAUSAL MULTIVIEW SIMULATION

Cosmos3-Nano-Sim-Auto

Distilled causal variant of Cosmos3-Nano-Transfer-Auto, published by NVIDIA on October 20, 2026, for closed-loop driving simulation. It produces synchronized multiview video and LiDAR from controls and text. It supports up to eleven cameras. The ability to respond to controls during a sequence distinguishes it from the bidirectional generator, but it does not constitute an autonomous driving policy.

21 OCT 2026 RESEARCH IN SURGICAL ROBOTICS

Cosmos3-H-Embodied-Nano

Research checkpoint published on October 20, 2026, for surgical robotics. A single model offers three modes: direct action-to-video dynamics, inverse video-to-action dynamics, and a policy conditioned on image, text, and history. It derives from Cosmos3-Nano, with a Qwen3-VL reasoning tower and fine-tuning of the generative tower. It is not three independent models nor an authorized medical device.

21 OCT 2026 INDUSTRIAL SIM-TO-REAL EDITING

CDF Sim2Real SMT Solderlight

NVIDIA image-editing model, version 1.0.0 released on October 20, 2026. It transforms synthetic PCB component crops rendered in Omniverse into an optical-inspection lighting style. It derives from Qwen-Image-Edit with a fused LoRA fine-tune: the package includes the complete pipeline, not a separate adapter. Users supply the image; the English instruction is fixed, not a free-form prompt.

20 OCT 2026 MULTIMODAL DOCUMENT RETRIEVAL

Nemotron 3.5 Embed 1B BF16

NVIDIA embedding model with 1.55B parameters, released on October 20, 2026. It vectorizes text documents, page images or both, with text-only queries. Distilled from Embed 8B, it freezes the Nemotron 3 Embed 1B text tower and adds visual retrieval. Its vector has 2,048 dimensions; it does not automatically share the 8B model’s space.

20 OCT 2026 MULTIMODAL DOCUMENT RETRIEVAL

Nemotron 3.5 Embed 8B BF16

NVIDIA embedding model with 8.38B parameters, released on October 20, 2026. It vectorizes text, page images or both for retrieval with text queries. It freezes the Nemotron 3 Embed 8B tower and adds vision through Pixtral. It retains that predecessor’s text embedding space; it produces 4,096-dimensional vectors, not conversational responses.

20 OCT 2026 MULTIMODAL DOCUMENT RERANKING

Nemotron 3.5 Rerank 8B BF16

NVIDIA reranker with 8.38B parameters, released on October 20, 2026. It scores each text-query and document pair; documents may contain text, a page image or both. It complements initial retrieval by reranking candidates; it does not generate responses or embeddings. It combines a Pixtral vision encoder with a bidirectional Ministral3 tower.

20 OCT 2026 STRUCTURED OCR WITH 0.631B PARAMETERS

Nemotron 3.5 Parse

NVIDIA model released on October 17, 2026 to turn page images into structured text, semantic classes, boxes and reading order. It combines C-RADIOv4-SO400M with a ten-layer mBART decoder and 630,800,496 unique parameters. It accepts an RGB image and a task prompt; PDFs must be converted into page images. It does not generate images or replace a general-purpose assistant.

18 OCT 2026 4B RECURRENT REASONING

Nemotron Labs Dory 1.0 4B

NVIDIA text model released on October 16, 2026, with 4 billion parameters and a recurrent architecture. It reuses a block of layers to adjust compute per token without switching checkpoints. English and a validated 256K-token context are documented. It does not generate images, audio or video.

17 OCT 2026

02 / EVALUATIONS

Published results

Figures are shown with the context reported by their source. A provider result is not an independent comparison and does not replace your own evaluation.

Model	Benchmark	Result	Metric
NVIDIA Nemotron 3 Ultra	Coste en SWE-bench	Up to −30 %	Published cost per completed task
Nemotron 3.5 Bolt E4B	BFCL-v3	BF16: 68.51% · Q4_K_M: 67.31%	Higher is better
Nemotron 3.5 Bolt E4B	ToolSandbox	BF16: 68.77% · Q4_K_M: 74.13%	Higher is better
Nemotron 3.5 Bolt E4B	RULER (64K)	BF16: 86.99% · Q4_K_M: 85.76%	Higher is better
Cosmos3-Nano-Sim-Transfer	PAI-Bench-C TransferBench DOVER ↑	10,175942	Score
Cosmos3-Nano-Sim-Transfer	PAI-Bench-C TransferBench segmentation mIoU ↑	0,73776	Score
Cosmos3-Nano-Sim-Transfer	PAI-Bench-C TransferBench depth SI-RMSE ↓	0,619426	Score
Cosmos3-Nano-Sim-Bimanual	PAI-Bench-G I2V Video Quality ↑	73,9099	Score
Cosmos3-Nano-Sim-Bimanual	PAI-Bench-G I2V ATE ↓	4,7072	Score
Cosmos3-Nano-Transfer-Auto	Cosmos3 Driving T2V CFG3 FID ↓	45,5	Score
Cosmos3-Nano-Transfer-Auto	Cosmos3 Driving T2V CFG3 FVD-I3D ↓	488	Score
Cosmos3-Nano-Transfer-Auto	Cosmos3 Driving I2V CFG3 FID ↓	27,8	Score
Cosmos3-Nano-Transfer-Auto	Cosmos3 Driving I2V CFG3 FVD-I3D ↓	238	Score
Cosmos3-Nano-Sim-Auto	Cosmos3 Driving T2V CFG1 FID ↓	48,6	Score
Cosmos3-Nano-Sim-Auto	Cosmos3 Driving T2V CFG1 FVD-I3D ↓	480	Score
Cosmos3-Nano-Sim-Auto	Cosmos3 Driving I2V CFG1 FID ↓	33,6	Score
Cosmos3-Nano-Sim-Auto	Cosmos3 Driving I2V CFG1 FVD-I3D ↓	325	Score
Cosmos3-H-Embodied-Nano	Cosmos3-H held-out 640 forward PSNR ↑	26,1 dB	Score
Cosmos3-H-Embodied-Nano	Cosmos3-H held-out 640 inverse action MSE ↓	0,009	Score
Cosmos3-H-Embodied-Nano	Cosmos3-H held-out 640 policy mean action MSE ↓	0,052	Score
Cosmos3-H-Embodied-Nano	Cosmos3-H held-out 640 policy best-of-4 action MSE ↓	0,033	Score
Nemotron 3.5 Parse	ParseBench Overall (triple-pass union)	75,20	Higher is better
Nemotron 3.5 Parse	ParseBench Tables (triple-pass union)	84,67	Higher is better
Nemotron 3.5 Parse	ParseBench Charts (triple-pass union)	64,40	Higher is better
Nemotron 3.5 Parse	ParseBench Content Faithfulness (triple-pass union)	87,96	Higher is better
Nemotron 3.5 Parse	ParseBench Semantic Formatting (triple-pass union)	68,28	Higher is better
Nemotron 3.5 Parse	ParseBench Visual Grounding (triple-pass union)	70,67	Higher is better
Nemotron 3.5 Parse	IndicVisionBench OCR ANLS-char	83,73	Higher is better
Nemotron 3.5 Parse	IndicVisionBench OCR CER	16,00	Lower is better
Nemotron 3.5 Parse	MOSCAR Character Bag F1	99,51	Higher is better
Nemotron 3.5 Parse	MOSCAR CER	1,35	Lower is better
Nemotron Labs Dory 1.0 4B	SciCode	34,94	BF16 score · 2 loops
Nemotron Labs Dory 1.0 4B	SWE-bench Verified	59,5	BF16 score · 2 loops
Nemotron Labs Dory 1.0 4B	Terminal-Bench 2.1 (Terminus 2)	18,1	BF16 score · 2 loops
Nemotron Labs Dory 1.0 4B	Tau3-bench (Banking)	12,89	BF16 score · 2 loops
Nemotron Labs Dory 1.0 4B	GDPval-AA v2 (NVIDIA anchors)	946	BF16 Elo · 2 loops
Nemotron Labs Dory 1.0 4B	GPQA Diamond (sin herramientas)	72,41	BF16 score · 2 loops
Nemotron Labs Dory 1.0 4B	HLE (texto, sin herramientas)	10,89	BF16 score · 2 loops
Nemotron Labs Dory 1.0 4B	IFBench	77,94	BF16 score · 2 loops
Nemotron Labs Dory 1.0 4B	AA-LCR (NVIDIA, Qwen3 judge)	71	BF16 score · 2 loops
Nemotron Labs Dory 1.0 4B	AA-Omniscience (accuracy)	10,08	BF16 score · 2 loops

03 / METHOD

How to read the directory

An organization, a product, and a model are not the same unit. Inferama separates them to avoid attributing capabilities or commercial terms to the wrong item.

Laboratory

Entity that develops or publishes the model and maintains its technical and safety documentation.

Model

Identifiable version with limits, modalities, and behavior that may change between releases.

Access channel

API, application, associated cloud, or commercial plan; each channel may have different pricing, retention, and limits.

Source and date

Every claim must retain the official page consulted and the verification date.

ANALYSIS

Related analysis

ANALISIS

Cosmos3-Nano-Sim-Transfer: Capabilities, Limits, and Real Deployment Costs

Technical analysis of NVIDIA's model for converting simulation controls into RGB video. We evaluate its architecture, published benchmarks, hardware requirements, and the physical limitations that prevent its use as a certified simulator.

22 Oct 2026 NOTICIA

NVIDIA releases Nemotron 3.5 Rerank and Embed, alongside CDF Sim2Real SMT Solderlight

Four NVIDIA models were made public on October 20, 2026: two embedding systems, a multimodal reranker, and an image editing model for board inspection. All operate under the OpenMDW 1.1 license and require self-hosted infrastructure.

21 Oct 2026 NOTICIA

NVIDIA releases five Cosmos3-Nano variants for simulation and robotics

On October 20, 2026, NVIDIA made five models from the Cosmos3-Nano family publicly available. These 16-billion-parameter checkpoints cover use cases ranging from multiview driving simulation to surgical robotics, and are available for direct download on Hugging Face and GitHub.

21 Oct 2026 ANALISIS

AA-LCR v1.1: Methodology, Metrics, and Limits of Long-Context Reasoning Evaluation

A technical audit of the AA-LCR benchmark: how 100 questions are constructed over 234 documents, the role of the Qwen3 judge model in scoring, and the execution conditions that ensure comparability between language models.

20 Oct 2026 ANALISIS

Cost in SWE-bench: how to calculate the price per resolved issue without hiding failures, retries, or evaluation

A cost figure per resolved issue is useful only if it reveals everything that happened before the patch was obtained: failed attempts, token consumption, stopping rules, selection among runs, and evaluation resources. This analysis proposes a reproducible scorecard for reading and comparing SWE-bench results, with particular attention to SWE-bench Verified.

22 Sep 2026

04 / SOURCES

Traceability