
Moonshot AI's Kimi K3 Drops 2.8 Trillion Parameters to Beat OpenAI and Anthropic
Moonshot AI's 2.8T parameter Kimi K3 launches with frontier-level agentic performance, a 1M token context window, and open weights on the way
20 stories

Moonshot AI's 2.8T parameter Kimi K3 launches with frontier-level agentic performance, a 1M token context window, and open weights on the way

Zyphra's ZUNA 1.1 upgrades its open EEG foundation model with variable-length inputs, realistic dropout training, and a no-code browser playground for brain signal reconstruction.

Cohere's North platform becomes the AI backbone of the University of Toronto, marking a symbolic homecoming for the startup's three founders

Cognition launches 'Devin for Startups,' offering VC-backed early-stage teams up to $65K in credits and matching grants to integrate the AI coding agent into their engineering workflow.

Screenpipe hits 20K GitHub stars and joins YC S26, turning your local screen and audio history into a live memory layer for AI agents

Google Cloud adds Parallel Web Systems as a native grounding provider on Gemini Enterprise Agent Platform, giving agents a dedicated search index built for agentic workloads

Gemini now lets you set up a persistent digital avatar once and place yourself in any AI-generated image or video without re-uploading a selfie each time.
How to decide whether a machine-generated change should influence the next run
H Company's fully managed computer-use agent API is now in open beta, powered by Holo3.1, the #1 model on OSWorld-Verified at 80.4%

Google open-sources GNM Head, a scan-based parametric 3D head model with 636 controls for identity, expression, and anatomy — free for commercial use.

Manus adds a /typst-pdf-maker skill that turns a single prompt into a professionally typeset PDF, no LaTeX or design tools required.

StackBlitz open-sources Bolt Slides, a React-based presentation engine that lets any AI agent build fully interactive, web-native decks from a single prompt.
microagi raises a record $55M seed round to teach factory robots by filming human workers — no robots or models required

Final2x v4.0 ships a new PyTorch backend with custom model support and Nvidia 50-series GPU acceleration, all for free

DeepStream 9.1 ships 13 agentic skills that let coding agents like Claude Code build full multi-camera 3D tracking pipelines from a plain-English prompt

Sakana AI integrates NVIDIA's Nemotron open model family into its Fugu multi-agent orchestrator, creating a feedback loop that could redefine how open models get deployed in production.

Google Research proves mathematically that diffusion model creativity is a side effect of regularization, not magic — and releases code to explore it

GitHub's Copilot coding agent can now be assigned Jira tickets, stream progress live, and iterate on pull requests without leaving Jira

Claude Code artifacts can now call your MCP connectors at view time, letting each viewer pull their own live data without the creator re-running anything.

Zed v1.11 ships dedicated staged/unstaged change views, a smarter Git Graph, and a symbols picker preview pane