Inkling 2 Mini: Thinking Machines' looped 26B model on a 32GB Mac
Thinking Machines has released Inkling 2 Mini, an Apache 2.0 looped MoE model that needs about 20GB of memory and runs locally on 32GB Macs.
Read articleInsights
Open-weight models, open-source AI projects and running AI on your own hardware.
Thinking Machines has released Inkling 2 Mini, an Apache 2.0 looped MoE model that needs about 20GB of memory and runs locally on 32GB Macs.
Read articleCloudflare has published Apache 2.0 weights for Clairic (27B) and Clairic-flash (9B), two models that turn documents into typed, scored fields.
Read articleOVHcloud's first open-weight model is a 6.8B Mamba-attention hybrid for short, tool-using tasks, but for now it only runs on OVH's forked runtimes.
Read articleTencent has released Hy4 flash, a 220B mixture-of-experts model with 12B active parameters, and 4-bit builds already run on 128GB machines.
Read articleEdge0-145B spreads its experts across GPU memory, system RAM and SSD so a 145B coding model runs on one consumer GPU, but its licence restricts hosting.
Read articleHigurashi's Apache 2.0 Haru-4B listens and speaks at once, runs on a 12 GB GPU or Apple silicon, and ships as a base model for research.
Read articleMBZUAI's IFM has released PAN 2 under Apache 2.0. It's a 29B world model that turns actions into simulated video, and it needs data-centre GPUs to run.
Read articleRunway's Praxis-1 robot control model is out under Apache 2.0, with weights that need an 80 GB GPU and a short list of known failure modes.
Read articlellama.cpp and ggml reached v1.0.0 on 13 November 2026, adding semantic versioning, FP8 KV caches and llama-server security fixes that change some defaults.
Read articleZyphra's Apache 2.0 ZONOS3-Mini runs speech synthesis with 100M active parameters on local hardware, though te reo Māori and phoneme control are missing.
Read articlePreferred Networks has released open weights for four translation models covering 53 languages, with GGUF and MLX builds small enough to run on a laptop.
Read articleIncept5's open-source inferd serves many AI agents from one Apple Silicon Mac, and an outside test shows what a smaller Mac can do.
Read articleRicursive's Ripple-65B Preview is an Apache 2.0 dense coding model that already runs in llama.cpp and vLLM, but its benchmark scores are self-reported.
Read articleBaidu's 280B-parameter ERNIE 5.5 Flash Base has open weights, and its licence bars use in the EU, the UK and South Korea.
Read article