Hancom is supplying its AI Agent technology to LG AI Research's ChatEXAONE platform as it moves to target the public and private sectors.
DeepSeek is permanently cutting the price of its flagship AI model by 75%.
It has been confirmed that a phenomenon called 'Constraint Decay' occurs, in which LLM agents' performance degrades sharply as structural constraints increase during backend code generation.
GitHub announced new Copilot personal plans featuring usage-based quotas.
PapersWithCode has updated its platform with multi-metric support, external paper registration, and paper lineage display features.
Introducing PICO, which is optimized for the human visual system and dramatically improves compression ratio and speed compared to existing codecs.
Anthropic has unveiled business workflow skill packs for AI agents.
Figure AI's F.03 robot sorted 250,000 packages over 200 hours, demonstrating human-level efficiency.
A bug in llama.cpp causes the KV cache of the Qwen3.6-35B-A3B model to be dropped, significantly degrading inference speed.
When using Claude Code, certain actions can invalidate the prompt cache, causing costs to spike by 12.5x.
Cumulative spending by frontier AI companies significantly exceeds cumulative revenue, indicating that the industry as a whole remains unprofitable.
Anthropic's new model 'Mythos' appears set to be integrated into Claude Code and Claude Security.
This analyzes various AI hardware technologies and corporate strategies aimed at solving the memory bandwidth problem, the core bottleneck in LLM inference.
Due to the absence of governance for AI agents, security risks are serious, including cases where corporate databases have been deleted.
Anthropic is preparing to introduce a security-specialized model, **Mythos 1**, into Claude Code and Claude Security.
The llama.cpp server now supports native tool-calling capabilities, including file manipulation and shell command execution.
An MLX port has been implemented that allows Cohere's 218B MoE model, Command A+, to run on Apple Silicon.
This analyzes the architecture of WordDetectorNet, which uses per-pixel distance regression and DBSCAN instead of an anchor-based approach.
llama.cpp now supports NVIDIA FP4 quantization and Multi-Token Prediction (MTP) features.
The CC 2.1.146 update introduces deterministic multi-subagent orchestration and workflow tools.
Cisco forecast that AI demand will spread beyond hyperscalers to enterprises and sovereign clouds, ushering in a networking-centric infrastructure cycle.
Apex-Testing, which evaluates AI agents' coding abilities using real GitHub repositories, has been updated.
DeepSeek has confirmed the 75% discounted price for the V4-Pro API as the permanent official price.
Native MTP support for Qwen3.6 MLX models has been added to the distributed inference tool exo.