AI BriefingPyTorchKR news
The latest AI and developer news from PyTorchKR, with short summaries.
Visit website →PickPrime Agent Released
Prime Intellect has released an open-source, self-improving coding agent based on an IPython kernel.
PickCloudflare Open-Sources Internal AI OS
Cloudflare has released its internal AI agent workspace, Cloudflare OS v2, under the Apache 2.0 license.
Cursor Releases MoE Kernel for NVL72
Cursor released a fused MoE communication and computation kernel for NVL72, reporting a 1.41x throughput improvement.
Liquid AI Releases LFM2.5
Liquid AI released a 2.6B agent model capable of running on mobile phones and CPUs.
DiffusionGemma Achieves 1,500 Tokens/s
DiffusionGemma adapts Gemma 4 into a discrete diffusion model, achieving 1,500 TPS on H100.
DeepSeek V4-Flash Released
DeepSeek has released the official version of V4-Flash, which supports a 1 million token context window.
MSIT Releases Manual on 21 LLM Threats
The Ministry of Science and ICT (MSIT) and the Korea Internet & Security Agency (KISA) have released a diagnostic and response manual covering 21 security threats to LLM systems.
Bento Releases AI Agent Presentation Tool
Bento has released a single-HTML presentation format with built-in editing and presentation capabilities.
x402, Agent Payment Standard
x402 integrates per-request payments for AI agents' APIs and MCP tools into the HTTP request flow.
PickInkling-Small Released
Thinking Machines Lab released an open-weight model one-quarter the size of Inkling.
AMD Releases Instella-MoE
AMD has released a fully open 16B MoE model trained on ROCm.
Marble Releases Elementary Learning Concept Graph
Marble has released 1,590 elementary learning concepts as a prerequisite relationship graph.
MiniMax releases H3 weights
MiniMax has released the weights for its 33B H3 model, which jointly generates video and stereo audio, but Korea is excluded from the license.
QM Releases AI Agent for Teams
QM has released an open-source AI agent harness that isolates workspaces per team member and chat room.
image-blaster, 3D Environment Generation
image-blaster converts a single image into a 3D environment, objects, and sound effects.
Google's Research Agent That Links Evidence
A Google research team proposed ScientistOne, which connects every claim in autonomously generated research papers to evidence.
SimpleMem Cuts Memory Tokens by 30x
SimpleMem compresses long-term memory to boost LoCoMo performance while cutting inference tokens by up to 30x.
OpenAI Unveils Codex Security
OpenAI has released a security agent that understands repository context to find, verify, and patch vulnerabilities.
darwin-skill, Agent Skill Optimization
darwin-skill 2.0 iteratively improves agent skills through evaluation, verification, and rollback.
AI Papers to Watch This Week
A roundup of 10 AI/ML papers covering real-world agent evaluation and token efficiency.