
Microsoft AI releases MAI-Transcribe-2-Streaming and MAI-Voice-2.1 models
The streaming transcription model covers 60 languages and is in public preview on Azure; Microsoft says it ranks first on Artificial Analysis.

Google's post lists a 1M-token output limit and $2/$10 per million token pricing; broad availability has no date yet.

The streaming transcription model covers 60 languages and is in public preview on Azure; Microsoft says it ranks first on Artificial Analysis.
The deal is reported at about $7 billion over five years with roughly 30% upfront; neither company has confirmed it.
The company says the modular system solved test problems up to 9,980 variables; benchmarks are its own and first shipments are expected in Q4.
Reuters relays a New York Post report; the FTC has not published a statement and the scope is not public.
The hand has 13 degrees of freedom from identical actuators and dense tactile sensors; the company dropped the pinky to cut cost and failure points.

Cognition reports up to 4.8x total token throughput versus GB200 NVL72 on its SWE-2 inference workload; figures are company-measured.

The system combines self-play reinforcement learning with test-time search over hidden pieces; the 20-game series against Pim Niemeijer included four draws.
The White House fact sheet covers official communications only; existing regulations, contracts and statutory definitions are unchanged.
Results for the fiscal quarter ended September 3 were filed with the SEC; AI memory demand drives the jump.
OpenAI's report describes up to 16,000 attempts from over 4,000 users in July; it attributes a core cluster to associates, not necessarily the company.
TRANSCEND-T2D-1 reported A1C drops of 1.7–2.0 points and up to 16.8% weight loss at 40 weeks; the paper is indexed on PubMed.