Open Source
-
Infrastructure & Chips
Amazon opens TorchNeuron so unmodified PyTorch code can run natively on its Trainium AI chips
Amazon has opened TorchNeuron, software that lets existing PyTorch code run on AWS Trainium chips after a configuration change instead of a rewrite, to all developers as a public beta. It plans to release the tool as ope…
-
Open Source
Tencent Hunyuan open-sources Hy-Memory 2.0, an agent memory stack that pairs a 30B memory-writing model with an SDK and a cross-tool desktop app
Tencent's Hunyuan team released Hy-Memory 2.0 as open source on October 15 and 16, 2026, opening three layers at once: the Hy-Memory-30B-A3B model for writing long-term memories, the hy-memory engine and Python SDK, and …
-
New Models
OpenMOSS releases MOSS-Image, a 5B open-weight text-to-image model aimed at dense text rendering and layout control
The OpenMOSS team released MOSS-Image-Base and MOSS-Image-SFT on October 16, 2026: a 5-billion-parameter diffusion transformer trained from scratch, published with Apache 2.0 weights, that renders long English and Chines…
-
New Models
Mistral introduces Timestral, a family of time-series foundation models up to 209M parameters
Mistral on October 15, 2026 released Timestral in four sizes from 5M to 209M parameters, saying the largest leads GIFT-Eval, TIME and fev-bench on probabilistic multivariate forecasting while running efficiently on CPUs.
-
Research
Microsoft-led study in Nature shows a lightweight ML correction can double AI forecast skill three to four weeks out
Researchers from Microsoft, MIT, ECMWF and other institutions published probabilistic bias correction in Nature on October 14, 2026. The open-source method learns a forecast model's habitual errors, doubled the weak week…
-
Open Source
Hugging Face releases Llama Windows, an open-source tray app that brings llama.cpp models to Windows 11
Hugging Face has launched Llama Windows, a free, MIT-licensed app that lives in the Windows 11 notification area, pulls open models from the Hugging Face Hub and serves them on the PC through llama.cpp with an OpenAI-com…
-
New Models
Tencent opens Hy4 preview for million-token coding and agent workflows
Tencent releases Hy4 preview with Apache 2.0 weights, 1M-token context, vLLM/SGLang deployment and a coding-focused 770B MoE architecture.
-
New Models
AMALIA opens a 9B European Portuguese model with a 32K context and local deployment path
AMALIA releases a 9B open European Portuguese model with 32K context, SFT+DPO training, public evaluation resources and local vLLM serving.