IonQ and the University of Cambridge open a quantum innovation centre at the Cavendish Laboratory
The centre sits in the Ray Dolby Centre; the announcement is a partnership opening rather than a new hardware result.
Meta releases Muse Wonder and Muse Wonder Full, its first frontier models for agents
Muse Wonder is on the Meta Model API and Muse Code; the Full model is subscription-only. Features include Active Memory and Active Focus.
FDA launches Rare Disease Framework Initiative to tailor drug reviews to small populations
A public meeting is set for October 28 and comments are open through December 11; changes to guidance and rules may follow.
Artificial Analysis launches AA-Robotics, testing language models that directly drive a robot arm
GPT-6 Astra leads with 69.4 across eight tabletop tasks, but models took about 30 minutes per task versus 20 seconds for a person.
Intel unveils D270, an air-cooled inference GPU with 1.3 PFLOPS FP8 and LPDDR5X memory
Shown at the OCP Global Summit, the Crescent Island-based card is sampling; Intel has given no general availability date or price.
Microsoft details Maia 300 inference chip, plans multi-gigawatt deployments from 2027
Paper finds subliminal learning can pass capabilities and backdoors between models via unrelated data
Google Cloud previews Gemini agent coworkers with their own email, calendar and Drive
Microsoft AI finishes pre-training MAI-Base-2, a 2.8-trillion-parameter mixture-of-experts model
It has 80 billion active parameters and trained on 16,000 GB200 GPUs for 38 days; post-training continues and no release date is set.
Arrowhead's RNAi drug ARO-MAPT cuts spinal-fluid tau by up to 73% in healthy volunteers
Interim Phase 1/2a data show effects lasting 90 days after the last dose; these are biomarker results, not Alzheimer's efficacy data.
SpaceXSI releases Grok Voice 3.0, which tops Artificial Analysis's speech-to-speech index
It scores 86.9 against 83.2 for GPT-Live-1, though OpenAI keeps a narrow lead on τ-Voice; API price stays $0.08 a minute.
DeepRSI benchmark gives AI agents 48 hours to improve real ML research tasks
Hillclimb's 15 tasks are scored by private held-out verifiers; Claude Fable 5.1 leads at 1548 Elo ahead of GPT-6 Astra at 1232.
Super Micro-linked contractor pleads guilty to diverting Nvidia AI servers to China
Anthropic expands Claude Code Projects, which coordinate parallel cloud threads on one goal
FDA approves Teva's Weltruza, a once-monthly subcutaneous olanzapine for schizophrenia
Anthropic tells investors Q3 revenue reached $16.8 billion, up 45% from Q2, reports say
FT, CNBC and Axios report adjusted profitability and positive cash flow ahead of a possible IPO; Anthropic has not published the figures.
Bristol Myers says actinium-225 therapy RYZ101 meets Phase 3 progression-free survival goal
The ACTION-1 interim analysis in neuroendocrine tumors after Lu-177 therapy disclosed no hazard ratio or medians; full data await a medical meeting.
NEURA Robotics introduces Neuron AI, a robot foundation model that combines vision, touch, sound and memory
The German company says it runs on five of its robot types; it published no benchmark scores and plans a live laundry demo at GTC Berlin.