GLM-5V-Turbo is an agent model that integrates multimodal perception, tool use, and end-to-end verification.
On SpatialBench, GPT-5.5 and Opus 4.7 got faster but accuracy barely moved.
Cursor's sale to xAI showed that the independence of AI coding apps has hit its limit.
Per-GPU KV cache locality determined LLM serving cost and TTFT.
Anthropic is closing in on raising $50 billion at a $900 billion valuation.
xAI unveiled Grok 4.3, boosting both benchmark scores and cost efficiency.
Claude Security has moved into public beta for Claude Enterprise.
Adding inference-time review to tool-calling agents reduced errors before execution.
Mozilla has taken a negative position on Chrome's Prompt API, citing interoperability risks.
Codex CLI 0.128.0 adds /goal, which repeats execution until the goal is achieved.
AISI assessed GPT-5.5's cyber capabilities as similar to Claude Mythos.
A build method for using CUDA and ROCm simultaneously with llama.cpp on Windows was shared.
Open-weight Kimi K2.6 outperformed Claude and GPT-5.5 in an AI coding challenge.
OpenClaw has disclosed the technical measures and process it introduced to strengthen the security and safety of its AI platform.
Throttle Meter reads local session files to display Claude Code's 5-hour rolling usage in real time.
On 2 nodes with 32 AMD MI50s, Kimi K2.6 int4 achieved 9.7 tok/s output and 263 tok/s input.
Interaction vulnerabilities were revealed in a network of over 100 AI agents.
Andrew Kelley said LLM-assisted PRs are easily distinguished from human mistakes.
An MLX 4-bit conversion of Mistral Medium 3.5 128B has been released.
SDK 2026.2 raises RNGD throughput by an average of 74.9% and strengthens deployment automation.
On a single RTX 3090, Qwen3.6-27B achieved a 218K context and stable tool calling.
On an RTX 5080 16GB, Qwen3.6-35B-A3B hit 30 t/s at 128K and 89 t/s for new tokens.
NVIDIA presents OpenClaw and NemoClaw as a security and governance solution for long-running AI agents.
In an 18GB M3 Pro bench, Gemma4:e4b ranked first with 72%, while Qwen3.5:9b scored 68%.