Articles Tagged “Open Source AI”
40 articles found
Holo4 Open-Weight Agents: Screens, Code and MCP in One
H Company’s Holo4 comes in 27B and 35B-A3B sizes, with open weights in four formats. See how one agent model handles GUIs, code, MCP and APIs.
Google AI Hits 300 Languages, Covering 86% of People
Google says its products now work in 300+ languages for 7 billion people, backed by open speech datasets covering 109 Indian and 27 African languages.
Methane-Mapping AI Spots 50% More Plumes From Space
Google and NASA JPL released MAPL-EMIT, a vision transformer that found 23,000 extra methane plumes and detects 50% more than human analysts.
IBM Granite Time Series Forecaster Goes Apache 2.0
IBM's 385M-parameter Granite PatchTST-FM-r2 leads permissively licensed zero-shot forecasters on GIFT-Eval, dual-licensed under Apache 2.0 and OpenMDW 1.0.
K2 Horizon Ships Six Open Models With Training Data
IFM's K2 Horizon releases six Apache 2.0 models from 0.9B to 375B parameters, publishing training data, code and logs alongside the weights.
Nvidia to Buy Hugging Face in a $12.93 Billion Deal
Nvidia agreed to acquire Hugging Face for $12.93 billion, its second-largest deal ever, and says the open model hub will stay open to everyone.
WebGPU Kernels Make Local AI in the Browser 2.57x Faster
Hugging Face published 207 WebGPU kernels as a JavaScript library, reporting a 2.57x geometric-mean speedup over ORT WebGPU on an Apple M4 GPU.
Edge AI Model Zoo Logs 837 Reproducible NPU Tests
EdgeFirst's new model zoo publishes 837 validation sessions for YOLO models across NXP, Hailo, Jetson, Qualcomm and Apple Neural Engine hardware.
TrueForge Open Source Agent Harness Cuts Costs 75%
TrueFoundry's MIT-licensed TrueForge harness completed Enterprise-Bench tasks for $2.90 against $11.80, a 75% cost cut driven by context engineering.
Embedding Models Guide: Dense vs Multi-Vector RAG
A practical 2026 guide to choosing embedding models for RAG: dense bi-encoders, multi-vector late interaction, and rerankers, with cost and memory math.
Kog Inference Engine Squeezes 30x From Existing GPUs
French startup Kog hit 3,000 tokens per second on standard AMD and NVIDIA datacenter GPUs, claiming a path to 30x faster LLM inference in software.
AI Agents Reproduced 2,226 ICML Papers in Just 19 Days
A Hugging Face community challenge used AI coding agents to audit 2,226 ICML 2026 papers, verifying claims across 6,816 public reproduction logbooks.
WeatherNext Cyclones Adds a Full Day of Forecast Lead
Google DeepMind published WeatherNext Cyclones in Nature and open-sourced the weights under Apache 2.0, adding over 24 hours of cyclone forecast lead time.
K-EXAONE 2.0 Ships 750B Open Weights Under Apache 2.0
LG AI Research released K-EXAONE 2.0, a 750-billion-parameter mixture-of-experts model with 37B active parameters, under a permissive Apache 2.0 license.
Cloudflare OS Open-Sources an Agent Workspace Platform
Cloudflare released Cloudflare OS as open source — an agent workspace with isolated code runtimes, a governance layer, and user-modifiable internal apps.
Local LLM Servers Compared: Ollama vs vLLM vs llama.cpp
Ollama, llama.cpp, vLLM and LM Studio all serve local models. Here's which one fits your hardware, from a 16GB laptop to a four-GPU workstation.
OlmoEarth Cuts Wildfire Risk Mapping to 30 Hours
Ai2's OlmoEarth Platform mapped North American wildfire risk in 30.5 hours instead of 4,737, a 155x speedup at fractions of a cent per square km.
NightRun Boots a Local LLM on a Pi 5 With No OS
NightRun is a Rust UEFI application that boots straight into a local LLM on Raspberry Pi 5 and x86 PCs, skipping the operating system entirely.
Real World VoiceEQ Benchmarks the Human Side of Voice AI
Hume AI and Hugging Face open a voice AI benchmark built on over one million human ratings, covering 40+ models and 60+ metrics of speech quality.
Fireworks AI Raises $1.5B for Faster Model Inference
Fireworks AI closed a $1.505 billion Series D at a $17.5 billion valuation, a 4.4x step-up in nine months, betting on production model inference.
NVIDIA NeMo Automodel Fine-Tunes Any Diffusers Model
NVIDIA and Hugging Face shipped distributed fine-tuning for any Diffusers model — FLUX.1-dev LoRA hits 53.73 images per second on eight H100s.
SingGuard-NSFA Brings Open Guardrails to AI Agents
Ant Group open-sourced SingGuard-NSFA under Apache 2.0 — four guardrail models covering 185 threat scenarios across 133 languages, free to download.
Kimi K2.7 Code Brings Open Weights to GitHub Copilot
GitHub Copilot's first open-weight model, Kimi K2.7 Code, reached Business and Enterprise plans on July 7 — a 1T-parameter MoE with 32B active.
PyTorch 2.13 Brings FlexAttention to Apple Silicon
PyTorch 2.13 landed July 8 with FlexAttention on Apple Silicon — up to 12x faster attention on Mac GPUs and 4x lower memory for LM training.
Cohere Transcribe Arabic: Best Open Arabic Speech AI
Cohere Transcribe Arabic, released July 7 under Apache 2.0, hits a 25.87 word error rate — the top open-source Arabic ASR, beating Whisper Large V3.
Ollama Raises $65M to Power Local Open-Source AI
Ollama closed a $65M Series B led by Theory Ventures on July 9, growing to 8.9M monthly developers and a presence in 85% of the Fortune 500.
Liquid AI's Antidoom Open-Sources a Fix for Reasoning-Model Doom Loops
Liquid AI open-sourced Antidoom, a training method that stops reasoning models from looping on repeated tokens — cutting doom loops from over 20% to near 1%.
Exo Labs' local.ai Helps You Run Frontier AI on Your Own Hardware
Exo Labs launched local.ai, a free platform that matches the best AI model to your own hardware and shows when running locally beats paying per API token.
Leanstral 1.5: An Open-Source Lean 4 Model Brings Formal Proof to Real Code
Mistral's Leanstral 1.5 is an open-source Lean 4 theorem-proving model that saturates miniF2F and flags real bugs, making formal verification practical.
PP-OCRv6 Brings Tiny, 50-Language Open-Source OCR to Hugging Face
PaddlePaddle released PP-OCRv6 on June 22, 2026 — a family of tiny open-source OCR models covering 50 languages, now available on the Hugging Face Hub.
GLM-5.2 Open Weights Arrive as a Top Coding Model at a Fraction of the Cost
Z.ai released GLM-5.2 open weights under an MIT license on June 16, 2026 — an open-weight coding model that rivals the best closed systems on long-horizon benchmarks at roughly one-sixth the cost.
Google's Gemma 4 12B Brings Multimodal AI to a 16GB Laptop
Google DeepMind released Gemma 4 12B on June 3, 2026 — an open multimodal model that reads images and audio and runs on a 16GB laptop, free under Apache 2.0.
Hugging Face Hub 1.18 Adds an AI Skills Marketplace to Its CLI
The new huggingface-hub 1.18 release introduces an 'hf skills' command, surfacing a growing marketplace of reusable AI skills for the open-source community.
NVIDIA Cosmos 3: The First Fully Open Omni-Model for Physical AI
NVIDIA's Cosmos 3, launched early June 2026, is the first fully open omni-model for physical AI — reasoning first, then generating across modalities.
Hugging Face Turns the Hub Into Agent-First Infrastructure — Every Gradio Space Now Speaks Directly to AI Agents
Hugging Face shipped an /agents.md endpoint on every Gradio Space and elevated Kernels to a first-class repository type in May 2026 — making the open AI stack natively callable by AI coding agents.
NVIDIA's Ising Models Bring Open-Source AI to Quantum Computing
NVIDIA launched Ising today — the world's first open AI models for quantum error correction and calibration, delivering 2.5x faster performance and 3x higher accuracy.
Arcee Trinity-Large-Thinking: America's Most Powerful Open-Source Reasoning AI
Arcee's Trinity-Large-Thinking brings 400B-parameter open-source reasoning to enterprises under Apache 2.0 — the most capable US-made open-weight model ever released.
OpenClaw Becomes GitHub's Most-Starred Project as the Open-Source AI Agent Wave Crests
The open-source AI agent framework OpenClaw has surpassed 250,000 GitHub stars, overtaking React — marking a pivotal shift in how developers worldwide build and ship autonomous AI agents.
NVIDIA Open-Sources Cosmos World Models and Isaac GR00T N1 — Giving Robotics Developers a Free Physics Engine for AI
The Cosmos world foundation models and GR00T N1 humanoid robot model are now available on GitHub, enabling simulation-to-reality robotics development on Jetson edge hardware.
NVIDIA Debuts Nemotron 3 — Open Models With 4x Throughput and a Million-Token Context Window
NVIDIA's Nemotron 3 family ships three tiers of open models optimized for agentic AI, plus 3 trillion tokens of training data for the community.