LG's EXAONE 3.5 Hits 1 Million Monthly Downloads Running on 5GB
LG AI Research's 4-bit quantized 7.8B bilingual model shrinks a strong Korean-English instruction follower into roughly 5GB of weights with a 32K context window.
20 stories
LG AI Research's 4-bit quantized 7.8B bilingual model shrinks a strong Korean-English instruction follower into roughly 5GB of weights with a 32K context window.
Lightricks dropped a new in-context LoRA for LTX-2.5 that rebuilds fine detail in soft or upscaled video and tiles its way to 4K on a single GPU.
A new open-source AI workbench runs the full research loop, from literature review to experiments to write-up, with any frontier model and your own keys.
Jina AI's 0.6B multilingual reranker beats models 10x its size on BEIR by scoring documents jointly in one context window.
A 1B ternary language model built on redesigned DeepSeek R1 and Llama-3.2 blocks brings CPU-only inference with tool-calling and robotics tags.
Supermemory open-sourced its discontinued Company Brain product, a Slack teammate with persistent memory that runs on your own Cloudflare account.
MiniMax quietly rolled out M3.1-Flash-Preview inside MiniMax Code, a lighter sibling to M3 aimed at fast, cheap daily coding work with tunable reasoning effort.
A community LoRA teaches MiniMax-H3's video model to spin the camera around a subject in a seamless, loopable 360 degree orbit from a single photo.
A hackable Python project turns a Raspberry Pi and an e-ink panel into a living bird portrait powered by fully local audio classification.
An indie modder reimplemented NVIDIA's DLSS 5 Neural Rendering runtime from scratch for Radeon GPUs, and just cut its performance cost by 74% in a day.
A ternary-quantized Qwen3.8-27B derivative packs uncensored reasoning, a one-row 'thinking fix,' and multi-token prediction into a 7.66GB file.
Alibaba's flagship 32B dense reasoning model gets an official 4-bit AWQ build, cutting VRAM enough to fit on a single 24GB GPU with minor accuracy loss.
A 1.2B parameter open-source vision-language model that parses both digital PDFs and warped phone-camera document photos in one framework
A community port brings Mirai Labs' 2.4-bit trellis-compressed Qwen3.8-27B to llama.cpp, fitting 128K context on a 12 GB RTX 3090.
Higgsfield's MCP connector lets Claude drive a full motion design pipeline, from image generation to a finished, editable After Effects project.
ISTA's DASLab shrinks a 176.9B-parameter Qwen mixture-of-experts model to 29.6GB resident memory by pruning half its experts and quantizing the rest to 3.5 bits.
A tiny 5.3M parameter EfficientNet-B0 in timm keeps racking up nearly a million monthly downloads thanks to its modern RandAugment training recipe.
A new Claude Agent Skill called onetake generates one-take product launch films by rebuilding UIs from screenshots and scoring transition continuity with an oracle.
Rizzo Flow is an open-source, local-first alternative to TypeSafe's Jev that returns typed decisions from an LLM by reading answer-letter logits instead of generating tokens.
A 4.39M-parameter BERT sentence encoder producing 128-dimensional embeddings sits on Hugging Face as a lightweight fixture for testing sentence-transformers pipelines.