
Anthropic reports small AI models may be months away from self-replication on commercial IoT devices
In evaluations, a 9-billion-parameter model ran inference on 8 of 9 hardware types tested, the company said.
Meta has released Muse Wonder and Muse Wonder Full, its most capable models to date, the company said. The models deliver state-of-the-art performance on personal and professional agentic tasks, according to Meta.
Read the full story

In evaluations, a 9-billion-parameter model ran inference on 8 of 9 hardware types tested, the company said.
NVIDIA says its updated MGX 1.2 specification adds rack-scale blind-mate liquid cooling and supports rack deployments up to 250kW.
The Guardian reported that openings for companies like Cedar are arising as medical debt is expected to grow.
Disney's offer acts as a floor, and rival buyers can still bid more.
The Maia 200 is already serving production workloads, including Microsoft Copilot, and running OpenAI models, Microsoft said.
The filing records no crashes, injuries or fatalities, and TNW reports it is not yet an investigation.
The piece says most major AI labs have recently gone quiet on democracy itself.
Sycamore Labs chief executive Sri Viswanath said some customers are narrowing where agents can operate, with chief information officers wanting agents and model calls kept within their virtual private clouds.
Amazon Web Services said GPT-6 Sol and GPT-6 Luna, models from OpenAI, are now generally available on Amazon Bedrock. Amazon Web Services said Sol is built for complex tasks performed throughout the week, and Luna for focused, repeatable tasks at high volume. Both models are priced significantly lower than their GPT-5.6 predecessors, the company said. Amazon Web Services said Sol and Luna can be reached through global cross-Region inference from all five AWS Regions.
Some enterprise customers are running through cloud AI token budgets faster than planned, Helen O'Sullivan, an AI business development manager at Dell Technologies, told SiliconANGLE. O'Sullivan said some Bay Area customers are using up, in weeks, budgets meant to last months. SiliconANGLE reported that the strain is pushing some companies to weigh on-premises infrastructure and open-weight models alongside cloud frontier models. Tim Wood, sales director for the Northwest region at Intel Corp., said many enterprises already have CPUs in their data centers.
Fondation Inria, with NVIDIA's support, is launching a working group on open source advanced modeling and tool adoption, with findings to come in the spring of next year, the post said.
The blog says ReplaySSM is integrated into TensorRT-LLM, vLLM and FlashInfer.
Two pilot projects have already helped identify and fix over 1,000 potential vulnerabilities, according to GOV.UK.