llama.cpp has released v1.0.0, adopting a semantic versioning policy based on C API compatibility standards.
The release removes deprecated APIs and adds FP8 support, YaRN truncation controls, and performance improvements across CUDA, Metal, Vulkan, and other backends.
The free, MIT-licensed tool saves work to disk to survive context compaction and allows handoff between different AI coding agents.
Ace is an open-source tool that enables teams to work with AI agents in shared channels on their own machines, requiring Tailscale and Anthropic or OpenAI keys.
Unsloth updated Qwen3.8-27B GGUFs with Dynamic V3.1 for 5-10% accuracy gains and introduced new 1-bit quantization types for the 2.4T parameter model.
The FAA halted all departing flights at US airports, citing a telecommunications outage, shortly after expanding the $1 billion SMART AI system from Washington, DC to the entire country.
The update introduces proactive status updates, read-only thread sharing, and branching commands to reduce manual monitoring of parallel cloud sessions.
The system saved an average of 10 minutes and 80 gallons of fuel per flight during its initial deployment in the National Capital Region.
The release expands hardware support to include two-GPU setups and achieves 75.5 tok/s on a single RTX PRO 6000 card.
Google AI Pro and Ultra subscribers can now claim monthly Google Cloud credits and set up billing directly within AI Studio without leaving the workspace.
The stateless browser runs on Cloudflare Workers and supports Chrome DevTools Protocol for agent interaction.
The update reduces system prompts from ~8k to ~1k characters and uses plain text markers instead of JSON to ensure reliable performance on smaller local models like Gemma 12B.
Cloudflare identifies six pillars for the Modern Security Model to counter autonomous AI agents that chain stolen credentials for lateral movement.
The release uses an LLM to natively edit weights, outperforming Unsloth and EXL3 in KL divergence and top-1 accuracy at 128K context.
IsoKron research found that Kubernetes, GitHub, and Grafana MCP servers, as well as PostHog's API, can lose concurrent writes in Claude Opus sessions, with the Kubernetes server losing changes in 20 of 20 runs.
The framework scales group size quadratically with spatial compression to enable 32x compression while maintaining competitive performance on five text-to-image benchmarks.
Jan von Plato used Claude to digest 1,009 pages of Gödel's 1940s workbooks, yielding 663 theorems where about 400 were verified as stated in Coq.
Researchers demonstrate that frontier models can manipulate other models' weights to induce unauthorized behaviors, creating a defensive burden where attackers need only one successful payload while defenders must verify all changes.
The new framework captures telemetry from first-party and custom agents into BigQuery, enabling teams to track reasoning, control token costs, and evaluate accuracy before deployment.
The 4-bit version requires only 4 GB of VRAM and achieves approximately 13x real-time speed on an RTX 5090.
Gnomon is an open-source PostgreSQL extension that provides branch, diff, and merge capabilities for database data, allowing AI agents to safely modify production data with full auditability and undo functionality.
OpenAI dissolved its independent robotics team to rely on general-purpose models like GPT-6 Astra for robot control, citing a 74.8% success rate in simulation benchmarks.
Developed by Ai2 and partners, XYBench contains 8,115 questions to measure how well models recognize when a user's request is based on a misconception and steer them toward the underlying goal.
A technical guide details serving the 753B parameter GLM-5.3 NVFP4 model across two NVIDIA DGX Stations using vLLM, achieving a peak throughput of 607.8 tokens/sec.
AI Briefing