Search “NVIDIA”

NVIDIA releases on-device small model Nemotron-3.5-Bolt-E4B

NVIDIA releases the on-device small model Nemotron-3.5-Bolt-E4B. A report on October 23, 2026 states that the model can run locally on devices with 6GB of memory, targeting on-device devices such as phones; it is trained based on Gemma 4, supports vision, and Unsloth simultaneously provides the GGUF file and a guide link. NVIDIA says it targets scenarios such as local assistants, interactive characters, and robots that can converse and call tools. A follow-up report on October 24 states that the model has 8B total parameters, 4B effective parameters, and a 64K context, and is post-trained using only reinforcement learning based on Google Gemma 4 E4B; the Q4_K_M version occupies 3.1 to 4.3 GB of VRAM on RTX and can run on GeForce RTX, Jetson, and DGX Spark. The report also states that in NVIDIA's own testing, it leads Gemma 4 E4B, Qwen3.5-4B, and Nemotron 3 Nano 4B in 15 of 21 benchmarks, but Tau2-Bench is 55.1, trailing Qwen3.5-4B's 75.1.

View story:NVIDIA releases on-device small model Nemotron-3.5-Bolt-E4B

MiniMax thanks NVIDIA NeMo for day-0 fine-tuning support for M3.1-Flash

On October 23, 2026, MiniMax publicly thanked the NVIDIA NeMo team, expressing gratitude for providing day-0 fine-tuning support for MiniMax-M3.1-Flash. According to information officially released by MiniMax, the cited post by Huiying Li states that the model had already processed 200T+ tokens through Space Bunny before its public release, and that the weights are now open. The official statement also mentions that NeMo-AutoModel's day-0 fine-tuning support covers FSDP2, expert parallelism, pipeline parallelism, and context parallelism, can scale to 128 GPUs, and can directly load NVFP4 checkpoints. Currently, this event has only this one official fact, with no subsequent developments.

MiniMax thanks NVIDIA NeMo for day-0 fine-tuning support for M3.1-Flash

On October 23, 2026, MiniMax publicly thanked the NVIDIA NeMo team, expressing gratitude for providing day-0 fine-tuning support for MiniMax-M3.1-Flash. According to information officially released by MiniMax, the cited post by Huiying Li states that the model had already processed 200T+ tokens through Space Bunny before its public release, and that the weights are now open. The official statement also mentions that NeMo-AutoModel's day-0 fine-tuning support covers FSDP2, expert parallelism, pipeline parallelism, and context parallelism, can scale to 128 GPUs, and can directly load NVFP4 checkpoints. Currently, this event has only this one official fact, with no subsequent developments.

View story:MiniMax thanks NVIDIA NeMo for day-0 fine-tuning support for M3.1-Flash

NVIDIA releases AIOps platform Minerva

On October 22, 2026, NVIDIA released Minerva, an AIOps platform for AI factories. According to NVIDIA's official blog, the platform collects telemetry data from infrastructure and AI workloads, then uses dedicated AI models to turn the data into actionable insights; it builds a dynamic knowledge graph that maps GPUs, SuperNICs, switches, cables, and the logic of the jobs running on them, analyzes billions of data points per hour, learns normal behavior, detects unseen anomalies, and predicts failures. The platform includes a built-in AI assistant that can connect to external agent tools and supports troubleshooting cluster health and providing remediation suggestions in natural language. IREN, IBM, Jio, Lambda, Tesla, and others plan to deploy it, while Dell, Lenovo, and Supermicro will incorporate it into their AI factory operations product portfolios. On the same day, Unite.AI reported that Minerva includes a multi-agent AI assistant named Sherlock and provides an MCP server, and that IREN, Bitdeer AI, and others are already using it. There are currently no subsequent developments.

NVIDIA showcases full lineup of local AI hardware products at GTC Berlin

On October 21, 2026, NVIDIA showcased three local AI hardware lines—RTX Spark, DGX Spark, and DGX Station—at GTC Berlin, and announced multiple software updates. Official information shows that DGX Spark 64GB goes on sale on October 23; NVIDIA Sync will support clusters of up to eight DGX Sparks and will add Model Launcher this month; DGX Station for Windows is expected to launch in December; the Recursive Flint local coding model is released with open-source Apache 2.0 weights. A follow-up report on October 23 stated that DGX Spark officially supports eight-node clusters, with unified memory expandable to 1TB, and can locally run frontier models such as the full Z.ai GLM 5.3 (753B); official benchmarks show that Qwen3.8-27B (NVFP4) decoding throughput increases from about 12 tokens/s on a single node to 52 tokens/s on eight nodes, with eight nodes achieving a 4.43x decoding speedup, and nanochat pretraining achieving a 7.79x speedup with 97.4% scaling efficiency. NVIDIA Sync's eight-node cluster support will be available before the end of the year.

NVIDIA showcases its full line of local AI hardware products at GTC Berlin

On October 21, 2026, NVIDIA showcased three local AI hardware lines—RTX Spark, DGX Spark, and DGX Station—at GTC Berlin, and announced multiple software updates. Official information shows that DGX Spark 64GB goes on sale on October 23; NVIDIA Sync will support clusters of up to eight DGX Sparks and will add Model Launcher this month; DGX Station for Windows is expected to launch in December; the Recursive Flint local coding model is released with open-source Apache 2.0 weights. A follow-up report on October 23 stated that DGX Spark officially supports eight-node clusters, with unified memory expandable to 1TB, and can locally run frontier models such as the full Z.ai GLM 5.3 (753B); official benchmarks show that Qwen3.8-27B (NVFP4) decoding throughput increases from about 12 tokens/s on a single node to 52 tokens/s on eight nodes, with eight nodes achieving a 4.43x decoding speedup, and nanochat pretraining achieving a 7.79x speedup with 97.4% scaling efficiency. NVIDIA Sync's eight-node cluster support will be available before the end of the year.

View story:NVIDIA showcases full lineup of local AI hardware products at GTC Berlin

NVIDIA releases OpenShell reference architecture

On October 21, 2026, NVIDIA released the OpenShell reference architecture for building, isolating, observing, and enforcing secure agents on enterprise infrastructure and edge systems; the architecture is now available. The architecture uses Vera Rubin to run model inference, Vera CPU to host the agent harness, and BlueField-4 DPU to run Sentry for chip-level policy enforcement; Sentry can isolate out-of-bounds agents at millisecond latency. On the same day, NVIDIA released Sentry, a security software framework that monitors AI agents and enforces access policies within a hardware-isolated trust domain on the BlueField-4 DPU, without relying on host software; the framework is part of the OpenShell reference architecture announced at GTC Berlin and is planned to be opened early next year. Subsequently, Red Hat and Nvidia expanded their collaboration, planning to integrate the OpenShell reference architecture as a native capability into its AI platform, and to provide runtime security controls, model access governance, and hardware support for autonomous AI agents through Red Hat AI Factory with Nvidia.

NVIDIA releases OpenShell reference architecture

On October 21, 2026, NVIDIA released the OpenShell reference architecture for building, isolating, observing, and enforcing secure agents on enterprise infrastructure and edge systems; the architecture is now available. The architecture uses Vera Rubin to run model inference, Vera CPU to host the agent harness, and BlueField-4 DPU to run Sentry for chip-level policy enforcement; Sentry can isolate out-of-bounds agents at millisecond latency. On the same day, NVIDIA released Sentry, a security software framework that monitors AI agents and enforces access policies within a hardware-isolated trust domain on the BlueField-4 DPU, without relying on host software; the framework is part of the OpenShell reference architecture announced at GTC Berlin and is planned to be opened early next year. Subsequently, Red Hat and Nvidia expanded their collaboration, planning to integrate the OpenShell reference architecture as a native capability into its AI platform, and to provide runtime security controls, model access governance, and hardware support for autonomous AI agents through Red Hat AI Factory with Nvidia.

View story:NVIDIA releases OpenShell reference architecture

Nvidia unveils five-layer AI architecture

On October 21, 2026, Nvidia unveiled a five-layer AI architecture during its keynote at GTC Berlin 2026. NVIDIA founder and CEO Jensen Huang described AI as a five-layer structure consisting of energy, AI factories, models, and applications, and said the world has reached the starting point of a new industrial revolution. According to an on-site update published the same day on the NVIDIA Blog, Huang also mentioned that European companies including L'Oréal, Revolut, Spotify, Merck, Vodafone, Zara, Porsche, Nestlé, and HSBC are working with NVIDIA to advance AI, and said NVIDIA has invested nearly $10 billion in Europe's technology ecosystem and collaborates with more than 12,000 partners. The next day, The Elec reported that the architecture is called the "Five-Layer Cake," with the five layers being, in order: power and energy infrastructure; accelerated computing hardware; AI factory infrastructure (including neoclouds and communications networks); AI models covering domains such as language, biology, chemistry, physics, robotics, and animation; and an application layer that connects tokens to industries and enterprises. Huang called on Europe to build a complete five-layer architecture of its own and to shift from being an AI consumer to an AI producer. There are currently no subsequent developments.

View story:Nvidia unveils five-layer AI architecture at GTC Berlin

NVIDIA expands open-source Agent Toolkit and launches Agent Knowledge Foundation

On October 21, 2026, NVIDIA announced the expansion of its open-source Agent Toolkit and the launch of the Agent Knowledge Foundation. According to information published by the NVIDIA Newsroom that day, the expansion also adds the OpenShell secure runtime, the NemoClaw blueprint, and tools such as NeMo Relay and NeMo Compass. The Agent Knowledge Foundation includes CUDA-X libraries and the new Auto-Ontology library, and integrates the Nemotron Retriever and the Kumo family of open models, allowing enterprises to use it without migrating data or rebuilding existing workflows. The same source said that companies such as Palantir, Databricks, and Snowflake have already integrated the toolkit into their respective platforms. There are currently no subsequent independent developments.

NVIDIA releases time series forecasting model Kumo Forecast

NVIDIA has released the time series forecasting model Kumo Forecast (formerly NV-Tesseract Forecasting). The model uses MOMENT-1-large as its backbone, employing an encoder of approximately 340 million parameters plus a trainable forecasting head, with the encoder weights kept frozen. Its DARR mode retrieves patterns from historical data via k-nearest neighbors and fuses them with direct forecasting results; it also provides an interpretable framework that outputs lag and feature attributions by forecast step. The model is released under the Apache 2.0 license and supports input sequence lengths from 256 to 2048. There are currently no subsequent developments.

View story:NVIDIA releases Kumo Forecast time series forecasting model

Mistral AI and NVIDIA partner to develop AI-native engineering workflows

On October 21, 2026, Mistral AI announced a partnership with NVIDIA to bring AI into the field of industrial engineering. According to Testing Catalog, the collaboration includes faster simulation, open benchmarks and accompanying datasets for physical AI, and evaluation of agentic AI in CAE and EDA scenarios. In addition, Mistral's models will be integrated into the NVIDIA Omniverse libraries for GPU-accelerated 3D scene generation. The partnership is currently at the announcement stage, with no further progress disclosed.

View story:Mistral AI and NVIDIA partner to develop AI-native engineering workflows

NVIDIA releases AIOps platform Minerva

On October 22, 2026, NVIDIA released Minerva, an AIOps platform for AI factories. According to NVIDIA's official blog, the platform collects telemetry data from infrastructure and AI workloads, then uses dedicated AI models to turn the data into actionable insights; it builds a dynamic knowledge graph that maps GPUs, SuperNICs, switches, cables, and the logic of the jobs running on them, analyzes billions of data points per hour, learns normal behavior, detects unseen anomalies, and predicts failures. The platform includes a built-in AI assistant that can connect to external agent tools and supports troubleshooting cluster health and providing remediation suggestions in natural language. IREN, IBM, Jio, Lambda, Tesla, and others plan to deploy it, while Dell, Lenovo, and Supermicro will incorporate it into their AI factory operations product portfolios. On the same day, Unite.AI reported that Minerva includes a multi-agent AI assistant named Sherlock and provides an MCP server, and that IREN, Bitdeer AI, and others are already using it. There are currently no subsequent developments.

View story:NVIDIA releases AIOps platform Minerva

Aravind Srinivas shares JOLT post-training research in collaboration with NVIDIA

Aravind Srinivas shared JOLT, a post-training method proposed by Perplexity in collaboration with NVIDIA. The method lets the same model serve as both teacher and student, and trains the teacher to learn from rewards from its own attempts while penalizing drift away from the student's behavior. On the MATH benchmark, JOLT reaches accuracy comparable to GRPO with 6.5 times fewer training tokens; JOLT+ additionally provides outcome-based rewards to the student model and clearly surpasses GRPO on math, coding, terminal tasks, and tool-use benchmarks. Currently this research is being shared publicly by Aravind Srinivas, with no further follow-up progress yet.

Aravind Srinivas shares JOLT post-training research in collaboration with NVIDIA

Aravind Srinivas shared JOLT, a post-training method proposed by Perplexity in collaboration with NVIDIA. The method lets the same model serve as both teacher and student, and trains the teacher to learn from rewards from its own attempts while penalizing drift away from the student's behavior. On the MATH benchmark, JOLT reaches accuracy comparable to GRPO with 6.5 times fewer training tokens; JOLT+ additionally provides outcome-based rewards to the student model and clearly surpasses GRPO on math, coding, terminal tasks, and tool-use benchmarks. Currently this research is being shared publicly by Aravind Srinivas, with no further follow-up progress yet.

View story:Aravind Srinivas shares JOLT post-training research in collaboration with NVIDIA

NVIDIA Releases Relational Foundation Model Kumo Relational

On October 22, 2026, NVIDIA released Kumo Relational, a pretrained relational foundation model that can make predictions directly on enterprise databases in a single forward pass, without task-specific training, feature engineering, or table flattening. According to the same release, across 41 tasks on 15 databases—RelBench V1/V2, SAP SALT, and 4DBInfer—the model outperformed supervised relational deep learning models and gradient-boosted tree pipelines with expert feature engineering; on SAP SALT, its MRR was 8% higher than the AutoGluon ensemble. The model is available through NVIDIA's open-source GPU-native structured-data-models library or NVIDIA NIM, and supports databases with 500B+ rows. There are currently no subsequent independent development facts.

View story:NVIDIA Releases Kumo Relational, a Relational Foundation Model

Nvidia unveils five-layer AI architecture at GTC Berlin

On October 21, 2026, Nvidia unveiled a five-layer AI architecture during its keynote at GTC Berlin 2026. NVIDIA founder and CEO Jensen Huang described AI as a five-layer structure consisting of energy, AI factories, models, and applications, and said the world has reached the starting point of a new industrial revolution. According to an on-site update published the same day on the NVIDIA Blog, Huang also mentioned that European companies including L'Oréal, Revolut, Spotify, Merck, Vodafone, Zara, Porsche, Nestlé, and HSBC are working with NVIDIA to advance AI, and said NVIDIA has invested nearly $10 billion in Europe's technology ecosystem and collaborates with more than 12,000 partners. The next day, The Elec reported that the architecture is called the "Five-Layer Cake," with the five layers being, in order: power and energy infrastructure; accelerated computing hardware; AI factory infrastructure (including neoclouds and communications networks); AI models covering domains such as language, biology, chemistry, physics, robotics, and animation; and an application layer that connects tokens to industries and enterprises. Huang called on Europe to build a complete five-layer architecture of its own and to shift from being an AI consumer to an AI producer. There are currently no subsequent developments.

NVIDIA Releases Kumo Relational, a Relational Foundation Model

On October 22, 2026, NVIDIA released Kumo Relational, a pretrained relational foundation model that can make predictions directly on enterprise databases in a single forward pass, without task-specific training, feature engineering, or table flattening. According to the same release, across 41 tasks on 15 databases—RelBench V1/V2, SAP SALT, and 4DBInfer—the model outperformed supervised relational deep learning models and gradient-boosted tree pipelines with expert feature engineering; on SAP SALT, its MRR was 8% higher than the AutoGluon ensemble. The model is available through NVIDIA's open-source GPU-native structured-data-models library or NVIDIA NIM, and supports databases with 500B+ rows. There are currently no subsequent independent development facts.

NVIDIA releases Kumo Forecast time series forecasting model

NVIDIA has released the time series forecasting model Kumo Forecast (formerly NV-Tesseract Forecasting). The model uses MOMENT-1-large as its backbone, employing an encoder of approximately 340 million parameters plus a trainable forecasting head, with the encoder weights kept frozen. Its DARR mode retrieves patterns from historical data via k-nearest neighbors and fuses them with direct forecasting results; it also provides an interpretable framework that outputs lag and feature attributions by forecast step. The model is released under the Apache 2.0 license and supports input sequence lengths from 256 to 2048. There are currently no subsequent developments.

Mistral AI and NVIDIA partner to develop AI-native engineering workflows

On October 21, 2026, Mistral AI announced a partnership with NVIDIA to bring AI into the field of industrial engineering. According to Testing Catalog, the collaboration includes faster simulation, open benchmarks and accompanying datasets for physical AI, and evaluation of agentic AI in CAE and EDA scenarios. In addition, Mistral's models will be integrated into the NVIDIA Omniverse libraries for GPU-accelerated 3D scene generation. The partnership is currently at the announcement stage, with no further progress disclosed.