Open source
8 unrolled threads about open source, each one readable on a single page.
Google@GoogleWe just released Gemma 4 — our most intelligent open models to date. Built from the same world-class research as Gemini 3, Gemma 4 brings breakthrough intelligence directly to your own hardware for...
SummaryGoogle released Gemma 4, an open-source AI model family built on Gemini 3 research, featuring multiple sizes, 256K context window, function-calling for agents, and multimodal support under an Apache 2.0 license for commercial use.Apr 2, 20264 tweets♥ 20.2K
Stitch by Google@stitchbygoogleToday, we’re open-sourcing the draft specification for DESIGN.md, so it can be used across any tool or platform. We’re also adding new capabilities. DESIGN.md lets you easily export and import your d...
SummaryGoogle is open-sourcing DESIGN.md, a specification for exporting and importing design rules across tools and platforms, enabling AI agents to understand design intent and validate accessibility standards. The specification includes new features like a Tokens section for design tokens, an experimental Components section, and a CLI validator tool to help build a shared visual language for AI-assisted design.Apr 21, 20263 tweets♥ 18.2K
Superman@thesupermanmxChina open-sourced a peanut-sized OCR that parses entire 100-page PDFs in one shot.. It's called Unlimited-OCR. Only 3B params. Runs locally. Every other OCR tool chops your doc into pages and loses...
SummaryChina released Unlimited-OCR, a 3-billion-parameter open-source optical character recognition model that can process entire 100-page PDFs in a single pass with 32K context window, eliminating the page-by-page limitations of traditional OCR tools. The model achieves 93% accuracy on parsing benchmarks, runs entirely locally for free, and supports multiple languages, positioning it as a free alternative to expensive cloud-based OCR services like AWS Textract and Google Vision.Jul 19, 20262 tweets♥ 14.4KGoogle@GoogleToday we’re introducing Gemma 4 12B — our latest open model that brings advanced agentic reasoning, vision and audio directly to your laptop. It delivers performance nearing our larger Gemma models w...
SummaryGoogle introduces Gemma 4 12B, an open-source AI model that runs locally on 16GB VRAM with advanced reasoning, vision, and audio capabilities. The model uses a novel unified architecture that eliminates separate multimodal encoders, reducing memory usage and latency while delivering performance comparable to larger models.Jun 3, 20264 tweets♥ 9.3K
Google DeepMind@GoogleDeepMindMeet Gemma 4: our new family of open models you can run on your own hardware. Built for advanced reasoning and agentic workflows, we’re releasing them under an Apache 2.0 license. Here’s what’s new...
SummaryGoogle releases Gemma 4, a new family of open-source models in four sizes (31B dense, 26B MoE, E4B, E2B) under Apache 2.0 license, designed for advanced reasoning, agentic workflows, and local deployment on custom hardware with up to 256K context window. The models support tool use, multi-step task execution, and mobile deployment with vision and audio capabilities.Apr 2, 20264 tweets♥ 8.7K
Alexandr Wang@alexandr_wang1/ big announcement today: we will be releasing an open weight version of muse spark 1.2 soon. we also are releasing muse glimmer, a 30B agentic model with open weights under apache 2.0. muse glimmer...
SummaryA company is releasing Muse Glimmer, a 30 billion parameter open-weight agentic AI model that runs efficiently on 24GB of VRAM and can operate as a fully capable agent with planning, tool calls, and failure recovery. The model is available on Hugging Face and compatible with multiple inference frameworks.Aug 10, 20266 tweets♥ 9.5K
Kye Gomez (swarms)@KyeGomezBIntroducing OpenMythos An open-source, first-principles theoretical reconstruction of Claude Mythos, implemented in PyTorch. The architecture instantiates a looped transformer with a Mixture-of-Expe...
SummaryOpenMythos is an open-source PyTorch implementation of a Recurrent-Depth Transformer (RDT) architecture that applies a fixed transformer block iteratively with Mixture-of-Experts routing, enabling improved efficiency-performance tradeoffs through inference-time reasoning depth rather than parameter count. The architecture uses LTI-stable recurrent injection, sparse expert activation, and multi-latent attention to achieve comparable performance to larger models with significantly fewer parameters.Apr 19, 202610 tweets♥ 8.2K
Mistral AI@MistralAI