New open models combine large total parameter counts with low active parameters and very long context windows, targeting local and edge deployment. Architectures like Kolibri-1, Ling-3.1-flash and hybrid KV-cache designs shift the bottleneck from compute to memory.
Blockway releases a 32B preview model with a hybrid architecture. It combines 54 linear-attention layers, 17 compressed-sparse attention layers, one full-attention layer, and hashed n-gram memory in host RAM. Only 18 of 72 layers keep a KV cache, ena...
A user assembled a fully local inference setup with GLM 5.3 Flash on two DGX Sparks, running vLLM with two-way tensor parallelism. Reported figures: prefill at about 1,500 tokens/s and decode at about 40 tokens/s with a 100k-token context. Claude Cod...
Aleph Alpha publishes the Kolibri-1 card: 78 billion total parameters, 3.46 billion active, and a 1 million token context under Apache 2.0. AI-RADAR analysis shows the constraint shifting from compute to memory and KV cache, with direct implications ...
The Kyojin project runs two roughly 300-billion-parameter MoE models on a single 128 GB mini PC with AMD Strix Halo. Measured on a Ryzen AI Max+ 395, GLM-5.3-Flash and MiMo-V2.6-Flash reach up to 650 tokens/s prefill and 26 to 44 tokens/s decode. EXL...
Aleph Alpha releases Kolibri-1, an LLM with 78 billion parameters, 3.46 billion active per token, and up to 1 million tokens of context under Apache 2.0. The combination touches three constraints of self-hosted deployments: memory, compute, and licen...
As reported, Ling-3.1-flash combines 560 billion total parameters, 25 billion active per token, and a context window up to one million tokens, with a two-week free trial and then an open-source release. The profile points to a sparse architecture: lo...
