CorridorKey: When a VFX Artist Builds the Tool
Corridor Digital co-founder Niko Pueringer released CorridorKey, an open-source AI chroma keyer that handles motion blur, hair, and translucent edges at professional quality.
SD3.5-Flash Brings 4-Step Image Generation to Phones
SD3.5-Flash generates images in just 4 steps instead of the usual 30 to 50, making on-device AI image generation practical on smartphones and laptops with under 8GB of RAM.
Google Colab MCP Server Connects AI Agents to GPUs
Google has released an open-source MCP server for Google Colab, letting AI agents like Claude Code and Gemini CLI create notebooks, execute Python code, and access cloud GPUs directly.
Holotron-12B Ships Open-Source Computer Use at 2x Speed
H Company released Holotron-12B on March 17, an open-source computer use agent model that delivers more than 2x the throughput of its predecessor on a single H100 GPU.
Unsloth Studio Brings No-Code LLM Training to Desktop
Unsloth AI released Studio, a free local application that provides a visual interface for fine-tuning large language models while using 70% less VRAM than standard training methods.
Mamba-3 Open-Source SSM Rivals Transformers at 7x Speed
Researchers from CMU, Princeton, Together AI, and Cartesia AI published Mamba-3, a state-space model architecture that matches transformer performance while running inference 7x faster.
Mamba-3: Can SSMs Finally Challenge Transformers?
Mamba-3 matches transformer quality at 1.5B parameters while running inference 7x faster. Three technical innovations close the SSM capability gap.
Leanstral Proves AI Code Is Correct at 93% Less Cost
Mistral released Leanstral, the first open-source AI agent built for Lean 4 formal verification, proving AI-generated code meets its specifications at 93% lower cost.
Mistral Small 4 Ships 119B Open Multimodal Model
Mistral AI released Mistral Small 4, a 119B mixture-of-experts model that unifies reasoning, multimodal vision, and coding under a single Apache 2.0 license.
NVIDIA Nemotron Coalition Unites 8 Labs for Open AI
NVIDIA launched the Nemotron Coalition at GTC, uniting Black Forest Labs, Cursor, Mistral AI, Perplexity, and four other labs to co-develop open frontier AI models.
Open-Source Creative AI Catches Up in March 2026
Six open-source releases in two weeks just made a full creative pipeline possible without a single subscription Between March 1 and March 15, 2026, something happened that has never happened before in creative AI: open-source models matched or exceeded commercial tools across every major creative...
NVIDIA Nemotron 3 Super: Open-Source 120B Model
NVIDIA released Nemotron 3 Super on March 11, 2026, an open-source 120B-parameter hybrid Mamba-Transformer model that delivers 5x higher throughput than its predecessor for agentic AI workloads
Hume TADA: Open-Source TTS With Zero Hallucinations
Hume AI has released TADA (Text-Acoustic Dual-Alignment), an open-source text-to-speech model that eliminates a persistent problem in AI speech synthesis: hallucinated words
Fish Audio S2: Open-Source TTS Beats GPT-4o
Fish Audio released S2, an open-source text-to-speech model trained on over 10 million hours of audio data spanning 50 languages. The model beats GPT-4o-mini-tts on the EmergentTTS-Eval benchmark with an 81.88% win rate, and ships with full weights, fine-tuning code, and a streaming inference engine
IBM Granite 4.0 Speech Tops OpenASR Leaderboard
IBM released Granite 4.0 1B Speech on March 6, a compact multilingual speech model that now ranks first on the OpenASR Leaderboard for automatic speech recognition
Modular Diffusers: Composable AI Image Pipelines
Hugging Face launched Modular Diffusers on March 5, replacing the monolithic DiffusionPipeline with a composable block system
LTX-2.3: Open-Source AI Video With Desktop Editor
Lightricks released LTX-2.3 on March 5, 2026, a 22-billion-parameter open-source video generation model with native audio, portrait video support, and a companion desktop editor that runs the entire model locally. The release ships under Apache 2.0, making it free for commercial use and fine-tuning
Kiwi-Edit: Open-Source AI Video Editing with Text and Reference Images
Kiwi-Edit, a new open-source video editing framework from NUS ShowLab, launched on March 5, 2026, combining text instruction guidance with reference image control to handle both global and local video edits at 720p resolution
FireRed-Image-Edit 1.1: Xiaohongshu SOTA Open-Source Image Editor
Xiaohongshu's Intelligent Creation team released FireRed-Image-Edit 1.1 on March 3, 2026, achieving a composite score of 7.94 across five authoritative benchmarks and claiming the top spot among open-source image editing models
Qwen 3.5 Small: Edge AI Models Run on Phones
Alibaba's Qwen team just released Qwen 3.5 Small, a series of four open-source models ranging from 0.8B to 9B parameters that run on phones, IoT devices, and browser tabs. The smallest model operates on a seven-year-old Samsung Galaxy S10E at 12 tokens per second
GLM-5: China Ships a 744B Open-Source AI Model Trained on Huawei Chips
Zhipu AI released GLM-5 on February 13, 2026, a 744 billion parameter open-source model trained entirely on Huawei Ascend chips without a single NVIDIA GPU