A 31-million-parameter voice model runs offline in a retro room where a typewriter reads back whatever you write.
Nemotron 3.5 Transcribe handles 40 languages with 30B parameters and only about 4B active per step
Carbon-X adds DNA tokens to a 14B text model so researchers can paste a sequence and ask about…
Flora splits annotation into segments, captions and rewards, and Nvidia says it already labels 200,000 hours a day.
At GTC Berlin the company said eight linked Sparks can hold a terabyte of memory, enough to keep…
Clairic answers structured questions from documents in one pass, with no generated tokens and no reasoning step.
A new benchmark found 81 of 134 models would help with mass casualty requests under a thin pretext.
Muse Spark Open becomes the largest US open-weight release to date, with vision and audio baked in.
Microsoft and Nace both shipped small decision-scoring models that rank choices rather than generate prose.
Holo4 ships in two sizes and switches between screen control, code execution and tool calls without swapping models.
Fireworks AI makes its training API generally available for custom model work.
Google DeepMind says Gemma has hit one billion cumulative downloads with more than 100,000 community variants.