VoxCPM2 Open-Sources 2B-Param TTS With 30 Languages
OpenBMB released VoxCPM2, a 2 billion parameter text-to-speech model that runs on 8GB VRAM, supports 30 languages at 48kHz, and can design voices from natural language descriptions.
MiniMax Music 2.6 Adds AI Cover and Style Transfer
MiniMax released Music 2.6 on April 10, adding a Cover feature that extracts a song's melodic skeleton and lets creators rebuild everything around it.
llama.cpp Ships Multi-GPU Tensor Parallelism
llama.cpp b8738 adds backend-agnostic tensor parallelism that enables large AI models to run across multiple GPUs without vendor-specific code.
GaussiAnimate Auto-Rigs 3D Gaussian Splatting
Researchers released GaussiAnimate, a framework that automatically rigs and animates 3D Gaussian Splatting assets with 17.3% quality improvement.
MegaStyle Trains FLUX on 1.4M Styled Images
Researchers from Tongji University, Tencent, and five other institutions released MegaStyle, a 1.4-million image dataset for style transfer alongside a FLUX-based model.
LPM 1.0 Generates Real-Time Character Video
A team of 23 researchers released LPM 1.0, a 17-billion parameter Diffusion Transformer that generates real-time character video from audio input.
Safetensors Joins PyTorch Foundation as Open Project
Hugging Face contributes Safetensors to the PyTorch Foundation, making the de facto AI model storage format a vendor-neutral, community-governed project.
Google Open-Sources Scion Multi-Agent Orchestrator
Google Cloud Platform has open-sourced Scion, an experimental orchestration testbed that runs multiple AI coding agents as isolated, concurrent container processes.
GLM-5.1 Ships Open-Weight 754B Model That Tops SWE-Bench Pro
Z.AI released GLM-5.1, a 754-billion-parameter open-weight model that claims the top spot on SWE-Bench Pro with a score of 58.4, surpassing Claude Opus 4.6 and GPT-5.4.
Meta Plans Open-Source Mango Multimedia AI Generator
Meta confirmed it will release open-source versions of its upcoming Mango multimedia generator and Avocado LLM, extending its open-weight strategy to next-gen frontier models.
The Creator's Guide to Running AI Locally in 2026
In April 2026, a creator with a $1,200 desktop PC and a 12GB graphics card can generate publication-quality images in under 10 seconds, produce video clips, run a conversational AI, and compose music without sending a single byte to the cloud.
Netflix Open-Sources VOID for Physics-Aware Video Removal
Netflix has released VOID (Video Object and Interaction Deletion), its first public AI model. The open-source tool removes objects from video while preserving physically plausible interactions.
DeepSeek V4 Will Run Exclusively on Huawei Chips
DeepSeek V4 will run exclusively on Huawei Ascend 950PR processors, marking a milestone in China's push to build advanced AI without American chip technology.
ComfyUI Adds Wan2.7 Video Generation via Partner Nodes
Alibaba Wan2.7 video generation model is now available directly in ComfyUI through Partner Nodes, bringing comprehensive video creation capabilities to the popular node-based workflow tool.
Tencent OmniWeaving Unifies 7 Video Gen Tasks
Tencent has released OmniWeaving, a unified video generation model that handles seven distinct tasks from text-to-video to reasoning-augmented generation, with publicly available weights and code.
Gemma 4 Rewrites the Open Source AI Playbook
Google released Gemma 4, a family of four open multimodal models under Apache 2.0. The 31B dense model scores 89.2% on AIME 2026, quadrupling Gemma 3. For creators and developers building on open models, this changes everything.
Sony Woosh: Open Source Sound Effects AI Model
Sony AI has released Woosh, an open source sound effects foundation model supporting text-to-audio and video-to-audio generation.
Open Source Audio AI Closes the Quality Gap
The first week of April 2026 delivered three significant open-source audio AI models that directly challenge paid alternatives in voice synthesis, sound effects, and multilingual speech.
Liquid AI LFM2.5-350M Runs AI Agents in 81MB on Mobile
Liquid AI releases LFM2.5-350M, a 350M parameter open-weight model that runs AI agents on mobile GPUs in just 81MB with 32k context window.
Arcee Ships Open Reasoning Model at 96% Less Cost
Arcee AI releases Trinity-Large-Thinking under Apache 2.0, ranking #2 on PinchBench behind Claude Opus 4.6 at just $0.90 per million output tokens.
OmniVoice: Open Source TTS Covers 600 Languages
OmniVoice is a zero-shot text-to-speech model supporting 600+ languages with voice cloning, an Apache 2.0 license, and inference 40x faster than real-time.
Alibaba Wan2.7-Image Unifies AI Generation and Editing
Alibaba released Wan2.7-Image, a unified AI model that handles both image generation and editing in one system, with fine-grained personalization, color control, and a Pro variant capable of 4K output.
Falcon Perception: 0.6B Vision Model Beats Meta SAM 3
Technology Innovation Institute releases Falcon Perception, a compact 0.6B parameter vision model that beats Meta SAM 3 across six key benchmarks while running on consumer hardware.
PrismML Open-Sources Bonsai 8B: A True 1-Bit LLM
PrismML emerged from stealth and open-sourced Bonsai 8B, a true 1-bit large language model that fits 8.2 billion parameters into just 1.15 GB of memory, roughly 14x more compressed than comparable 16-bit models.