The latest in AI, clustered and ranked. Repeated hype gets pushed down so the actual signal stays up top.
OpenAI has expanded its GPT-6 lineup with Sol and Luna, two new models optimized for cost efficiency and faster performance.
GPT-6 Sol and Luna leverage the same training methodologies as the flagship GPT-6 Astra. These models are designed to provide high-level capabilities in coding, reasoning, and computer use while reducing operational costs for professional workflows.
A Pentagon investigation found that overreliance on AI systems contributed to a missile strike on an Iranian school.
NVIDIA introduced Topograph to optimize GPU workload scheduling by accounting for network topology.
The new llm-typesafe plugin for the LLM CLI tool enables structured, type-safe outputs using the Jev model.
Microsoft has dismantled EvilTokens, an AI-assisted platform used to compromise over 12,000 user accounts.
OpenAI updated GPT-6 prompt caching with improved hit rates, new diagnostics, and explicit breakpoints.
Pentagon report links AI overreliance to a missile strike incident on a school in Iran.
Microsoft disrupts EvilTokens, an AI-assisted platform used for mass account compromise.
Qualcomm announces new smartphone chips capable of running 30B MoE models locally.
OpenAI releases GPT-6 Sol and Luna, offering lower pricing with negligible performance gains over predecessors.
Interview with John Platt regarding the application of AI in scientific research and automation.
OpenAI released GPT-6 Sol and Luna, two new models optimized for different capability and cost profiles.
NVIDIA Topograph provides topology-aware workload scheduling to optimize GPU cluster performance.
A library for enforcing type safety in LLM outputs.
PsyAgentBench evaluates LLM agents on classic psychology experiments to distinguish between bias and response patterns.
PRQuant introduces Permutation Residual Quantization to improve low-bit inference accuracy by addressing weight outliers.
Summarize-Judge-Refine decouples multimodal understanding from policy learning for modular content moderation.
Performance and pricing benchmark analysis for Claude Opus 5.5.
Guide on optimizing ROS 2 node performance using NVIDIA Isaac ROS and GPU acceleration.
Tutorial on using Amazon SageMaker concurrency sweeps to benchmark and right-size generative AI endpoints.
Google open-sourced AX, a Kubernetes-style orchestrator for managing stateful autonomous AI agent workloads.
Guide on evaluating agent skill selection and instruction following using Strands Evals.
Overview of the Xiaomi MiMo-V2.6-Pro model, a new open-weights model trained for $3M.
![Thumbnail for [AINews] Xiaomi MiMo-V2.6-Pro 1T-A42B: the new top Open Weights model, trained for $3M](https://substackcdn.com/image/fetch/$s_!qfA0!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fpbs.substack.com%2Fmedia%2FHSxCs7wa0AANR91.jpg)
NVIDIA's SoL-Pi introduces harness mechanisms for coding agents, reducing token traffic by ~49% on EdgeBench.
SpaceXAI releases Grok 4.7, a larger base model with improved RL training at existing price points.
NVIDIA announces DLSS 5 with 3D-guided neural rendering and updates to the ACE and RTX development kits.
Commentary on the current state of AI marketing and recent industry security claims.
Overview of NVIDIA Confidential Computing for securing sensitive LLM inference workloads.
GitLab Duo Self-Hosted now supports model deployment via Microsoft Foundry on Azure.
Researchers are using a large language model to reconstruct missing text in ancient Greek papyrus fragments.
AstroForge plans to deploy a transformer-based model for autonomous control of a space probe.
Cisco Talos developed a framework to detect malware utilizing AI-driven autonomous command systems.
OpenAI's stated principles for third-party AI safety assessments.
Cloudflare adds support for private MCP server portals via Cloudflare Gateway.
Guide on using Amazon Bedrock Data Automation and Model Context Protocol for public sector data processing.
Explanation of choice versus score metrics in the Jev framework.
GPT-6 Sol and Luna models are now available on Amazon Bedrock.
Claude Opus 5.5 is now available on Amazon Bedrock for agentic and long-running tasks.
Anthropic releases Claude Opus 5.5.
Case study on using Amazon Bedrock AgentCore for automating Shopify app refreshes.
Podcast discussion covering RSI, US-China AI competition, and model scaling trends.
Trane Technologies reports using Amazon Bedrock AgentCore to optimize building diagnostic workflows.
Interview discussing AMD ROCm and the potential for AI agents in low-level hardware programming.
Podcast interview with the CEO of TypeSafe AI regarding System One models.
Overview of Jev's new System One decision models.
NVIDIA introduced multi-device inference support in TensorRT to distribute model workloads across multiple GPUs.
Xiaomi released MiMo-V2.6-Pro-RL, a multimodal vision-language model.
Product launch for Plane Agents, an AI-integrated project management tool.
A guide on evaluating AI agent performance through sequential tool calls and task completion metrics.
Same signal-first ranking, earlier dates.
Open archive