-
release · Hacker News · 3건Korean coverage · GeekNews · AI타임스
LFM2.5-24B-A2B: Built for fast on-device agents
Liquid AI released LFM2.5-24B-A2B, a mixture-of-experts model that activates only 2B of its 24B parameters per token. It targets fast agents running on-device, and both the base and post-trained models are available as open weights.
With only 2B active parameters per token and open weights, it is a candidate to download and benchmark for on-device tool-using agents where latency and data privacy matter.
-
release · Mistral
Introducing Timestral
Mistral introduced Timestral, a state-of-the-art probabilistic forecasting model family offered in four sizes from 5M to 209M parameters. It is described as CPU-efficient and ready for zero-shot use.
With sizes from 5M to 209M parameters and a CPU-efficiency pitch, you can test whether forecasting fits on existing CPU servers without GPUs, which changes inference cost and deployment design.
-
product · TechCrunch AI · 2건
Google’s Universal Cart shopping tool launches in the US
Google's Universal Cart, previewed at I/O in May, is now rolling out in the US across the Gemini app and Google Search. It lets users save products from multiple retailers in one unified cart, and items added via the Gemini app and Search AI Mode carry across Google platforms.
Developers building shopping agents or commerce integrations should check where their own cart and saved-product features now overlap with a cart built into Gemini and Search.
-
research · Hacker News
Simons Foundation launches $15M effort to help mathematicians navigate AI era
The Simons Foundation has launched a $15M effort to help mathematicians navigate the AI era. The news was shared on Hacker News.
Foundation money is now flowing into AI's role in math research, so developers building proof-assistant or mathematical reasoning tools can gauge researcher demand and possible collaborations.
-
release · Hugging Face
Granite Switch 4.1: Chat, Decisions, and Speech from One Endpoint
Hugging Face published Granite Switch 4.1. According to its title, it covers chat, decisions, and speech from a single endpoint.
Developers running separate models and endpoints for chat, decision logic, and speech can evaluate whether one endpoint could simplify their architecture and operations.
-
research · Google DeepMind
Accelerating the search for antimicrobial peptides with Google Antigravity
Google DeepMind describes using Google Antigravity for Science in the search for novel antimicrobial peptides. The stated aim is to accelerate that search.
It shows Google Antigravity for Science being applied to a real discovery pipeline, so teams automating scientific workflows can add it to the tools they compare.