#huggingface

New sanoTTS stack offers lightweight, multi‑language TTS for $3 microcontrollers with 11 voices and 337 KB int8 models, outperforming larger rivals.

Microsoft’s new VibeVoice-ASR-Streaming-7B delivers real‑time speech recognition, tackling speaker diarization in overlapping conversations for AI‑powered applications.

Explore H3-World, a new framework that turns language prompts into precise character and camera moves in video generation with only 0.199% trainable parameters.

Explore why Qwen 3.8’s advanced code‑generation can clash with real‑world workflows and how developers tweak quantization, caching, or hybrid models to balance power and practicality.

Explore Spark‑X2.5, a 4B/1.7B LLM with 20 trillion‑token training, Qwen‑level performance, and unique architecture—now on Hugging Face, ready for local deployment.

New SpeakoFlow Mini cleans dictations with 70.7% accuracy, outpacing GPT‑5.6 Luna and fine‑tuning a Qwen3.5 model for precise, offline speech‑to‑text edits.

New uncensored GGUF LLMs unveiled: “Heretic” and “Ultra Heretic” reduce refusal rates, 69B LongCat‑Flash‑Sparse handles 1M tokens, plus Qwen‑based models with vision and MoE support.

Explore how speculative decoding and n‑gram integration boost LLM inference speed, reduce latency, and cut costs—key insights for developers and AI ops teams.

OpenAI’s rapid executive exits, from COO to AI safety leaders, reveal a leadership shake‑up, yet the company’s strong revenue growth keeps it ahead of competitors.

GLM 5.3 weights released – compare its HumanEval, speed, and 256k context to rivals, plus insights on local‑host security use.