Catch today's product, model, and platform updates in a few minutes.
NVIDIA's early measurement data shows the Vera Rubin NVL72 delivers up to 30x higher per-megawatt throughput than the GB300 NVL72 on agentic workloads, with up to 35x lower cost per million tokens. These early results don't yet reflect Vera CPU tool-calling performance and are pending SemiAnalysis review.
Ox Alpha, a new AI model listed on OpenRouter as a "stealth model," was developed by an anonymous third party and is positioned as a reasoning model for coding and production workloads. It went live on OpenRouter for free. Its identity has drawn speculation, with some pointing to Chinese GLM models and others suspecting an unreleased version of Microsoft's MAI.
Inherent, an AI lab founded in London by Google DeepMind alumni, says its 27B-parameter AI agent Faraday outperforms Anthropic's Claude Opus 4.8 and OpenAI's GPT-5.5 at the task of replicating research papers. According to the arXiv paper and the company's research blog, Faraday is trained via long-horizon reinforcement learning using Inherent's proprietary Replica task suite (310 tasks drawn from 100 machine-learning and AI-for-science papers), replicating papers without access to the original plots. Faraday runs on a comparatively small Qwen 3.6 model with just 27B parameters, smaller than larger frontier models. The result remains a company self-report and has not been independently verified; Inherent chief scientist Edward Hughes said the key point lies in the way it was built.
Harvard Business School's HBS Foundry, an eight-week, $699 startup bootcamp for founders, introduces AI avatars of its instructors created by HeyGen, which provide personal feedback during practice pitches and board meetings while weekly live sessions with real instructors remain. Project director Katharina Rings said she initially envisioned a chatbot, but after releasing a trial version, students wanted a more guided experience. New York Times reporter Sarah Kessler tested it and said the AI version of Flybridge Capital co-founder Jeff Bussgang showed a noticeably frozen smile during her pitch; Bussgang admitted his digital copy is a bit "creepy" but said, "My students love it."
Meta Superintelligence Labs has released Muse Glimmer, an open-weight agentic model that anyone can download and run on their own hardware. The company open-sourced the model weights under the permissive Apache 2.0 license, and the 30-billion-parameter model is now available on Hugging Face as meta-models/Muse-Glimmer-30B. Muse Glimmer is optimized for always-on local agent workflows and is described as small enough to run on a Mac or PC with a single consumer GPU. The model card notes it is distilled from Muse Spark and includes a perception encoder, with 4-bit quantization shrinking it to under 20 GB so it fits within 24 GB or 32 GB VRAM envelopes, including on Nvidia RTX 5090 and Apple M4/M5 Max. Developers can get started immediately through the Ollama library by running "ollama run muse-glimmer," which enables fully local use without cloud or network access. No pricing or hardware-requirement changes accompany the release, and no limits beyond typical GPU memory constraints have been detailed. The Apache 2.0 license means builders can freely use, modify, and distribute the weights for commercial applications. Since the model is released as open weights, teams can deploy always-on local agents on their own infrastructure, with the main uncertainty being real-world throughput and memory performance on individual consumer-grade hardware setups.