AI BriefingPyTorchKR news
The latest AI and developer news from PyTorchKR, with short summaries.
Visit website →State of AI Report 2026: Frontier Big Three Landscape and Agent Security Incident Analysis
The 2026 State of AI Report provides a detailed analysis of the frontier big three landscape and agent security incidents.
AnySpeechRecognizer: Swift package supporting open models compatible with Apple Speech API
AnySpeechRecognizer, a Swift package that reimplements the Apple Speech API to enable the use of open models, and AnyDecisionModel for decision tasks have been released.
PickKOMOS Achieves SRCC of 0.900 in Korean Speech MOS Prediction
The Korean speech quality assessment model KOMOS has been released as open source, recording a higher correlation than UTMOS.
Baseten joey-1, 2B scale with P50 20ms, $0.01 per API token
Baseten unveiled joey-1, a 2B decision model, offering a price of $0.01 per 1M API input tokens.
NVIDIA Nemotron 3.5 Transcribe Released as Open Source
NVIDIA has released Nemotron 3.5 Transcribe, which supports 40 languages, as open source.
Weekly AI Paper Trends: Expanding Agent Evaluation Scope and Structured Learning
The summary introduces three major trends—expanding agent evaluation scope, structured learning, and refining data quality—along with 10 papers.
Gemini 4 Carbon Tops the AA Index with a Score of 59
Gemini 4 Carbon takes first place with an AA Index score of 59 and is prioritized for provision to cyber defense organizations.
Qwenputer embeds RISC-V CPU by converting GDN state to RAM
Qwenputer, an experimental model adding a RISC-V CPU to Qwen3.5-4B, is released, recording 97.5% accuracy on synthetic arithmetic benchmarks
Mirai's uzu achieves 142.3 tok/s speculative decoding with Qwen3.6 27B on M5 Max
uzu, a Rust inference engine for Apple Silicon, recorded a speed of 142.3 tok/s on the M5 Max.
OpenQodex Unveils Pre-Push AI Agent Code Review Tool
OpenQodex is an open-source tool where a separate AI session reviews code changes before pushing, achieving a 98% reproduction rate for seeded bugs.
Dharma AI Cuts Inference Costs by 77% for Qwen3-4B-Based Grading Model Using VocabX Technique
Reduced inference costs by 77% for a Qwen3-4B-based grading model through cost-weighted vocabulary selection.
JetBrains ThinkRail Supports Parallel Execution of Pi Agent Git Worktrees and Turn-Based Review
JetBrains released ThinkRail v0.2.0, an IDE with the Pi coding agent embedded.
Claude Code team doubles merges and cuts reverts by 40% with AI review system
Anthropic increased productivity while maintaining code quality through a six-layer inspection structure and convergence measures.
Runway Unveils Praxis-1, a Robot Policy Model Pretrained on Web Videos
Runway announced Praxis-1, a robot policy model pretrained on web videos, and previewed an open-weight release within months.
Meta RecIR: Single Codebase for GPU and TPU Training of PyTorch Recommendation Models
Meta has released the RecIR compiler in alpha version, enabling PyTorch recommendation models to be trained on both GPUs and TPUs.
tstack Releases 9 Skills for Automating Coding Agent PR Merging and Review Response
tstack, released by Theo Browne, is a set of 9 skills that enable coding agents to handle the process up to PR merging.
CopilotKit Open-Sources OpenDuet, a Self-Hostable Proactive Work Agent
CopilotKit has open-sourced OpenDuet, a self-hostable proactive work agent.
Liquid AI Unveils Six 2B-Scale LFM3 Models
Liquid AI has released six 2B-scale LFM3 models supporting execution on edge devices and multimodal processing.
Alibaba OpenGCUA, Python Harness Boosts OSWorld-v2 Average Score by 19.6 Points
The OSWorld-v2 average score rose from 45.3 to 64.9 points through harness changes alone, without model training.
IBM Unveils Granite Timeseries Ensemble R1, Combining Four Time Series Models
IBM has unveiled Granite Timeseries Ensemble R1, which combines four time series models.