
Stanford's OpenJarvis Runs Personal AI Agents 800x Cheaper Without the Cloud
Stanford's OpenJarvis is a free, open-source framework that runs full AI agents locally on your hardware, with no cloud required and a built-in learning loop
20 stories

Stanford's OpenJarvis is a free, open-source framework that runs full AI agents locally on your hardware, with no cloud required and a built-in learning loop

A community-built telecom LLM fine-tuned from Gemma 4 lands on Hugging Face, boosting context-grounded correctness by nearly ten points on standards QA.
Anything's new built-in Analytics dashboard gives app builders revenue tracking, retention, and bounce rate without a third-party tool

A new LoRA by ostris slashes Ideogram 4 inference from 20+ steps to just 2-4, using a novel non-distillation training technique still being refined.

NCCL 2.30.7 ships zero-SM collectives, MPS partition support, and major GIN networking enhancements for large-scale GPU training and inference

Zed v1.7 adds automatic context compaction to its native agent, letting long coding sessions run indefinitely without hitting model token limits

GPT-5.4 and Molecule.one's Maria AI ran 10,080 reactions to find a surprising additive that improved a key drug-making reaction for over 80% of tested substrates.
Midjourney's new Draft Mode for V8.1 generates 24 low-res images per prompt at half the GPU cost of a standard 4-image job

Google and 10 industry partners launch ARD, an open specification letting AI agents securely discover and verify tools, MCP servers, and other agents across the web.

qmd brings production-grade hybrid search — BM25, vector embeddings, and LLM reranking — to your local markdown files, with zero cloud dependencies

Odyssey closes a $310M Series B at a $1.45B valuation, betting that world models — AI that simulates physics and causality — are the next foundation model category after LLMs.

Ai2 releases MolmoMotion, an open 3D motion forecasting model that predicts where objects will move next from video frames and language instructions, beating all existing baselines on a new benchmark

xAI's Grok 4.3 lands on Amazon Bedrock with a 1M-token context window, configurable reasoning, and $1.25/M input pricing — but the real story runs deeper than a model listing.

Lightricks releases an IC-LoRA that adds photorealistic water to any dry video clip in 4.5 hours of H100 training on just 572 paired examples
v0's redesigned design mode adds a Figma-like toolbar, layers panel, drag handles, and a team annotation system directly on the live preview
Larksuite's official CLI brings 200+ commands and 26 AI Agent Skills to Lark/Feishu, letting agents send messages, manage calendars, and edit docs from the terminal

Replicate's Cog v0.20.0 renames cog predict to cog run, adds model refs, multi-source weights, and drops legacy schema generation entirely

The Pokémon Company and Kaggle launch a $240,000 two-stage AI competition to build the best Pokémon TCG playing agent, culminating in a live tournament in Japan.

Alibaba's LOGOS is a single 8B autoregressive model that handles proteins, molecules, materials, and reactions using one shared scientific grammar

NousResearch's hermes-agent-self-evolution uses GEPA and DSPy to let agents rewrite their own skills, prompts, and tool code — no GPU needed, ~$2–10 per run