Tag: Open Weights
-

Apple Harbor App Could Turn Your Mac Into a Private AI Server: What’s Reported So Far
Claris is reportedly testing Harbor, a Mac app that runs AI models privately and can pool several Macs. We sort the reports from the unknowns.
-

Qwen4-Max Costs $1.50 per Million Tokens, but Alibaba Won’t Let You Download It
Alibaba’s new flagship undercuts Claude Opus 5.5 on price, but you can’t download it, and most of the benchmark claims are Alibaba’s own.
-

Liquid AI LFM3 Fits in About 1 GB: How to Run This Small Open Model on Your Own Device
Liquid AI’s new LFM3 models are open-weight, tiny, and built to run on your own hardware. Here’s what the 1.7B version can do and how to try it.
-

Mistral Large 4 Open Weights Are Out: What the License Really Allows
Mistral just released a 1-trillion-parameter model you can download. Here’s the license catch, the API price, and who should self-host.
-

Kinference Explained: Buoyant’s Open Source Router That Aims to Speed Up vLLM on Kubernetes
Buoyant’s new open source Kinference sends each AI request to the vLLM server with the least work to do. Here’s what it claims and how to try it.
-

ERNIE 5.5 Flash Base Is Free to Download: What Baidu’s 280B Open Model Is Good For
Baidu just put a 280-billion-parameter model on Hugging Face. Here’s what you get, what’s missing, and who should bother.
-

Inkling 2 Small Is Live on Tinker, and Open Weights May Follow Next Week
Thinking Machines’ new 594B model is hosted-only today. Here’s what it costs, how it scores, and what to watch before the weights land.
-

Nemotron 3.5 Transcribe Is Free, but Can Your PC Run It?
Nvidia’s new open speech model transcribes 40 languages and labels speakers, but it needs a very big GPU. Here’s who it’s really for.
-

Liquid AI LFM3 Fits a Real AI Model in 1GB: What Runs on Your Phone
Liquid AI’s new LFM3 family puts six open models on phones and laptops, and the text model needs about 1GB. Here is what it can and can’t do.

