Daily AI News | NVIDIA releases Nemotron 3.5 Super VL and expands Bolt lineup
# Daily AI News | NVIDIA releases Nemotron 3.5 Super VL and expands Bolt lineup
## Lead Story
### NVIDIA releases Nemotron 3.5 Super VL and expands Bolt lineup
NVIDIA released Nemotron 3.5 Super VL, an open 120-billion-parameter model for specialized always-on AI agents. Super VL and the locally oriented Bolt 74B-A9B 24GB are available now, while the smaller Bolt E4B 4GB is scheduled for October 23.
**AI take:** Server and local deployments have different memory and latency needs. The separate variants give developers more deployment choices.
## More News
### Meta releases Muse Spark Open weights
Meta released weights for Muse Spark Open, a multimodal model aimed at coding and agentic tasks. The company describes 816 billion total parameters, 42 billion active parameters and a one-million-token context window; weights are distributed through platforms including Hugging Face, with API access from several providers.
**AI take:** Open weights create more deployment and customization options. Local feasibility still depends on memory, quantization and hardware.
### Black Forest Labs opens FLUX 3 Image Klein weights
Black Forest Labs announced open weights for FLUX 3 Image Klein, with base, distilled and several quantized versions. The company says the 5.6-billion-parameter image model retains native 4K generation and supports fine-tuning and local creative use.
**AI take:** The variants offer choices between customization, speed and memory use. Results still need testing on the user's own material.
### Anthropic begins giving METR employee-like evaluation access
Anthropic said it has begun giving METR researchers employee-like access to assess its models, safety practices and risk management. The company says METR will publish risk-assessment results and that it is also working with other external evaluators.
**AI take:** Deeper access can help independent researchers examine safety practices. The published methods and findings will determine what the assessment establishes.
### FSB urges financial institutions to strengthen firm-wide AI governance
The Financial Stability Board published AI adoption practices emphasizing firm-wide governance and board and senior-management oversight of alignment with strategy and risk appetite. The report warns that AI agents do not reason like humans and may interpret predefined goals in unexpected ways.
**AI take:** The recommendations extend oversight beyond initial deployment. Institutions need clear responsibility for goals, actions and exceptions.
### Adobe and NVIDIA expand creative-tool optimization work
Adobe and NVIDIA announced an expanded collaboration to optimize AI creation and production workflows in Photoshop, Premiere, Substance 3D and Brand Concierge. The work builds on their earlier strategic partnership and integrates NVIDIA accelerated-computing technology.
**AI take:** Creators will judge the work by time saved in specific tasks. The benefits need to arrive in concrete application releases.
### GeForce NOW adds DLSS 4.5 controls
NVIDIA added DLSS 4.5 controls to GeForce NOW for features such as super resolution, ray reconstruction and dynamic multi-frame generation in supported games. The settings are available in its Windows, macOS, Chrome OS and Steam OS apps, with options depending on membership tier, cloud hardware and game support.
**AI take:** The controls let users choose their balance of image quality and smoothness. Network conditions still affect the cloud-gaming experience.
### Google launches Universal Cart in the US
Google announced US availability of Universal Cart, which saves and tracks products found through Search and Gemini. It collects items from multiple retailers and can notify users about price drops or restocking, following its preview at the company's May developer conference.
**AI take:** A shared product list can reduce repeated searches. Purchase decisions still depend on the current price and the retailer's terms.
### Amazon's Bee introduces Frontier and Eclipse AI devices
Amazon-owned Bee introduced two personal AI devices: Frontier can be worn on the wrist or clipped to clothing, while Eclipse attaches magnetically to a smartphone. Both continue Bee's conversation-assistant approach, using AI to turn everyday discussions into summaries, tasks and information users can revisit.
**AI take:** The form factor affects recording habits. Buyers should also consider recording consent, subscription features and data handling.
### Bluesky expands Attie to websites and interactive experiences
Bluesky announced an expansion of Attie from custom social feeds to websites, games, quizzes and polls. The experiences run in the AT Protocol ecosystem, and creators can bring their own AI model or use open-weight options.
**AI take:** Connecting creation tools to an existing social network can ease distribution. Creators still need to test and maintain what they build.
### Mindlogic launches BAZE with access to about 80 AI models
Mindlogic launched BAZE, a personal platform offering about 80 generative AI models through one account, including text and video services. The reported basic plan costs 5,900 won per month for 3,000 credits, with additional credits available for purchase.
**AI take:** One interface makes comparing models easier. Actual cost still depends on how many credits each task consumes.
### Xenon launches GenA's full release and Android app
Xenon announced the full release of its GenA AI-agent portal and an Android app, adding to existing desktop and mobile web access. The platform brings together agents for work and everyday tasks, with an iOS app planned for later this month.
**AI take:** A common entry point reduces switching between services. Its value depends on how well the individual agents handle real tasks.
### Meitu's MVLAND adds marketing music-video generation
Meitu's MVLAND launched a marketing music-video feature for merchants, brand teams and creators. Users upload product material and choose a style to generate music and visuals for uses such as product previews, promotions and social-media content.
**AI take:** Combining music and visuals can reduce tool switching. Product details, material rights and output quality still need human checks.
### Zesty adds automated GPU fractioning decisions
Zesty announced Automated Fractional GPUs with support for NVIDIA's CUDA-native GPU fractioning. The company says it adjusts memory allocations, compute-sharing modes and workload placement as demand changes, within performance and availability policies set by teams.
**AI take:** GPU sharing works best when workloads can coexist. Evaluations need to track latency as well as utilization.
### TPIsoftware introduces AI appliances for enterprise workflows
TPIsoftware announced AI appliances preconfigured with OrientAI for contract review, HR, teaching preparation and enterprise AI governance. The company says its governance option manages agents centrally with role-based permissions, token-use limits and audit logs.
**AI take:** Preconfiguration can simplify deployment preparation. Enterprises still need to verify permissions, data flows and review responsibilities.
### Volvo Cars and NVIDIA expand automated-driving collaboration
Volvo Cars and NVIDIA announced a broader collaboration combining Volvo's vehicle-safety engineering with NVIDIA AI and accelerated computing for future cars. The work includes Hyperion sensor and compute architecture, Halos OS and Alpamayo Stack driving software.
**AI take:** The technology selection is an early step. Available driving features will depend on validation and integration into actual vehicles.
### RoboParty announces the RP1 humanoid research platform
RoboParty announced RP1, also called ROBOTO 01, for robotics research and embodied-AI development. The company says it opens body designs, resources for its ROMOMO joint actuators, low-level drivers, control algorithms, simulation and development tools for researchers to modify and test.
**AI take:** Access to engineering resources can aid reproducible experiments. Teams still need to check licenses, documentation and hardware availability.
### Rice Robotics launches TOMO and joins Hong Kong embodied-AI hub
Rice Robotics HK announced TOMO, a wheel-based humanoid robot primarily intended for education, and joined FLAIR's Embodied AI International Applications Hub as an early technology partner. The collaboration will train and validate humanoid robots and embodied AI in real environments, initially exploring education and manufacturing.
**AI take:** Real-world validation can reveal problems missed in a lab. Classroom use also needs workable supervision and a good experience for teachers and students.
### Swedish startup 120ft raises angel funding for robotic microfactories
Swedish startup 120ft launched with 576,000 euros from 22 angel investors to develop microfactories operated with humanoid robots. Its proposed service handles assembly, inspection and packing for brands, charging by finished product rather than requiring customers to buy robots.
**AI take:** Paying for output could change how small businesses adopt robotics. Reliable delivery still depends on quality control and production costs.
### Dogness introduces an AI pet feeder built with Google Gemini
Dogness announced a new AI pet feeder and technical cooperation agreements with Google Cloud. The company says the product uses Google Gemini to recognize behavior, monitor pets and issue alerts about unusual activity.
**AI take:** Behavior alerts can assist an owner's observations. They do not establish that the device can provide medical diagnoses.
### Crowded launches a financial MCP connector for ChatGPT and Claude
Crowded launched an MCP connector that lets nonprofits query live financial data and use supported workflows through natural-language requests in ChatGPT or Claude. The company says it retains Crowded's authentication and permissions, supporting tasks such as spending queries, fund-account organization and report preparation.
**AI take:** Live access can reduce manual exports. Financial actions still require checking the permissions and the actual result.
## Rumors
