
Groq Raises $350M From Nvidia After Selling Its Chip Soul
Groq banks $350M at a $3.5B valuation, half its former peak, as it completes the pivot from LPU chipmaker to Nvidia-powered inference cloud.
20 stories

Groq banks $350M at a $3.5B valuation, half its former peak, as it completes the pivot from LPU chipmaker to Nvidia-powered inference cloud.

A free open-source CLI grades your AI agent on 49 alignment tests across five failure modes, returning an A-F scorecard in under two minutes.

Remotion rebuilt its agent skills from scratch, splitting them into modular sub-skills and stripping out design opinions so agents produce cleaner React video code.

Nous Research turns Hermes agent profiles into a persistent roster of named bots that chat, run cron jobs, and message each other.

A UCLA and Allen AI team trained a humanoid to walk up and grab objects using only simulated data, hitting 73.3% real-world success on a Unitree G1.

GitHub's Blackbird code search team open-sourced a Rust crate that case-folds source code at over 45 GiB/s by deleting an early-exit branch.

A community NVFP4 quant shrinks the 27B uncensored Qwen3.8 to 20 GB, hitting 367.9 tok/s at concurrency 8 on four Blackwell cards.

GitHub Copilot code review now offers two selectable depths, letting you dial reasoning up for risky changes and down for trivial ones.

Zero is an open-source Go coding agent that runs in your terminal, works with 25+ model providers, and keeps sessions on your disk.

Diffusion Studio open-sourced a video editor and CLI that lets Claude Code, Cursor, and other coding agents write, cut, and generate video by talking to them.

An open-source Python toolkit bundles reinforcement learning, LLM forecasting, factor mining, and live trading into a single stock trading platform.

Prime Intellect ran 153 autonomous research runs across 18 frontier models on the nanoGPT optimizer speedrun, with the best closing 82% of the gap to a record built by dozens of humans over months.

A single 200-token proxy tool replaces thousands of verbose MCP definitions, letting Pi coding agents connect servers without draining context.

OpenBiliClaw is a local-first agent that builds a psychological profile of you and hunts content across 12 platforms.

Empero's tiny reasoning model squeezes a 2.4 trillion parameter teacher into a 1.3 GB file that runs on phones with a 262k context window.
Empero distilled a 2.4 trillion parameter teacher into a 4B model that runs on laptops with 3GB of RAM, with a big MMLU jump.

A community 4-bit quant of Qwen3.8-27B keeps the multi-token prediction head in BF16, hitting 112 tokens per second on a single Intel Arc Pro B70.
India's $1.5B sovereign AI unicorn takes its models, researchers, and live hackathons directly to university campuses nationwide

Sarvam AI's Kivi voice app will come pre-installed on HP laptops in India, bringing 22+ language voice control to millions of PCs

MiniMax released H3, a 33B open-weight video model that generates video with native stereo audio from mixed text, image, video, and audio inputs.