Kolibri-1, GLM 5.3 Flash, Qwen-Image-2.1-Turbo, and compact MoE/edge releases are redefining local and multimodal open-weight inference.
Nandakishor M. has released Vega, a physics-based typed decision model with 800 million parameters (a 4B variant is available), support for 73,000 tokens and images, and single-shot results that surpass many JEV benchmarks. The Test-Time Training arc...
Two open models lead the Artificial Analysis Cyber Index and outperform every single Anthropic model. The overtake is not just a ranking: it shifts the TCO calculation from API fees to on-premise infrastructure, reopening the debate on data sovereign...
Two open-source models hold the top positions on the Artificial Analysis Cyber Index, ahead of every Anthropic model. GLM 5.3 Flash leads; Mistral Large 4 also ranks better than its closed rivals. For those evaluating on-premise deployment, the shift...
The new accelerated checkpoint in the Qwen family generates and edits images in eight denoising steps. Open weights, a ready Diffusers pipeline and the same 7B visual architecture as Qwen-Image-2.1. Fewer iterations means faster inference, while the ...
Qwen-Image-2.1-Turbo's presence on Hugging Face adds another option for local image generation. The Turbo variant targets lower latency, a critical factor for self-hosted deployments with limited VRAM. The repository becomes a testbed for evaluating ...
An arXiv preprint describes LittleBit, an ultra-low-bit quantization approach based on latent factorization and quantization-aware training for very small models. The news matters for on-premise LLM operators: it is not just about saving space, but a...
The LiquidAI/d1-3B page on Hugging Face, shared without technical details, places the project in the compact LLM segment. The 3B label suggests a model aimed at local inference, self-hosted setups, and data sovereignty constraints. The article analys...
A 19,969-parameter, 81 KB FP32 model generates short stories with goal, problem, and resolution while running on CPU. MacroStories lowers the bound for coherent narrative generation and questions assumptions about local inference.
EmbeddingGemma 2 maps text and images into a 768-dimensional space. Ported to ruNNtime, a TypeScript WebGPU inference library, it enables photo search by description with inference entirely on the browser GPU. No upload, data stays on device: a concr...
TinyDecide is a 10-million-parameter model with a roughly 6 MB footprint, smaller than every entry on the Decision Index leaderboard. It runs in the browser, Node.js, Python, Rust, and on an ESP32. For edge and on-premise deployments, the signal is n...
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 researcher trained Apex-2, a decoder-only MoE with 3.87 billion total parameters and 1.45 billion active per token, from scratch on 86.5 billion tokens while scaling from one to two GH200 with DiLoCo. The base checkpoint reaches HumanEval+ parity w...
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 ...
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...
