A new open-source AI agent called Harness-1 outperforms GPT-5.4 with 73% accuracy using minimal training data, while Microsoft launches its own MAI model family, Apple offers free cloud AI access to indie developers, and OpenAI confidentially files for an IPO at a staggering $852 billion valuation.
Xiaomi and TileRT claim a major AI speed breakthrough, hitting over 1,000 tokens per second on a trillion-parameter model using just 8 commodity GPUs — roughly 15 times faster than ChatGPT and Claude — powered by FP4 quantization and speculative decoding, with an open-source model checkpoint already live on Hugging Face.
Nvidia-backed startup Generalist raises $400M at a $2B valuation in the physical AI race, even as experts and industry leaders warn that allowing a handful of companies to monopolize generalized robotics technology poses serious risks to humanity, prompting calls for open-source collaboration and government accountability.
Anthropic launches an open-source AI security tool on GitHub that uses Claude to autonomously detect and patch software vulnerabilities, running agents in isolated sandboxes through a full recon-to-patch pipeline with threat modeling, scanning, triage, and automated patch validation.
A new open-source CLI tool called 'Ollama Model Tester' is now available on GitHub, enabling users to run identical prompts across multiple local AI models simultaneously with zero external dependencies, making side-by-side model comparison faster and more efficient than ever.
NVIDIA launches Nemotron 3 Ultra, a massive 550B-parameter AI model delivering frontier reasoning capabilities at 30% lower cost and 5x higher throughput than comparable open models, powered by cutting-edge hybrid architecture and a novel multi-teacher training method, with fully open weights released for enterprise adoption.
PRIME-RL launches as a fully open-source reinforcement learning framework capable of scaling to 1000+ GPUs and trillion-parameter models, supporting major AI model families with asynchronous training, multimodal capabilities, and flexible multi-node deployment under the Apache 2.0 license.
Miso Labs launches MisoTTS, an open-source 8-billion-parameter text-to-speech model now live on Hugging Face, featuring voice cloning, highly emotive conversational speech, and built-in audio watermarking for safety.
A new open-source AI hiring agent launches on GitHub, capable of parsing resumes, analyzing GitHub activity, and delivering scored, explainable candidate evaluations using either local Ollama models or Google Gemini for flexible, privacy-conscious deployment.
Google DeepMind launches Gemma 4 12B, a multimodal AI model that runs locally on consumer laptops with just 16GB of VRAM, featuring a groundbreaking encoder-free architecture that processes vision and audio directly through its LLM backbone for lower latency and memory use, available now on Hugging Face and Kaggle under an Apache 2.0 license.