AI Infrastructure news and analysis on Web Pulse: 15590 curated articles tracking the latest AI Infrastructure developments, tools, and research, updated continuously from vetted sources.
OpenAI released gpt-realtime-2.1 and gpt-realtime-2.1-mini voice models on July 6, featuring a 25% reduction in p95 latency through caching improvements and adding reasoning and tool use to the mini model. The update eli…
Velocity, a startup founded by former ironSource and Unity executives Tal Shoham, Amir Shaked, and Nimrod Zuta, has raised $27 million to build an advertising layer for AI chat interfaces. The company applies mobile app …
Google Cloud has made AlphaEvolve, a Gemini-powered code-optimization agent, generally available on the Gemini Enterprise Agent Platform after a private preview. The tool searches candidate algorithms against user-suppli…
A developer builds an AI agent from scratch to demystify the core components—the loop, ReAct pattern, system prompts, message history, and tool calling—and explains why production agents require more than a naive loop, i…
AMD GPUs running SGLang Diffusion on ROCm achieve 1.5x to 6.3x speedups over Hugging Face Diffusers for image generation and editing tasks, as demonstrated on models including FLUX.1-dev, Qwen-Image, and Z-Image-Turbo. T…
Fable 5 and GPT-5.6 Soul can be combined in a multi-agent workflow where Fable 5 handles planning and review while GPT-5.6 Soul executes tasks, cutting inference costs by up to 10x without sacrificing quality. The orches…
OpenAI used its GPT-5.6 Soul model to autonomously post-train a smaller model called Luna, marking a concrete example of recursive self-improvement in AI. The larger model generated training data and evaluated outputs wi…
OpenAI has introduced GPT-5.6 with three model tiers—Soul, Terra, and Luna—optimized for intelligence, balanced performance, and speed respectively. Soul handles complex reasoning at higher cost, Terra serves as a genera…
XAI released Grok 4.5, a large language model offering near-frontier coding performance at $2 per million input tokens and $6 per million output tokens, roughly half the cost of comparable models. The model is optimized …
Tencent released Hunyuan-3, a 295-billion-parameter mixture-of-experts language model that activates only 21 billion parameters per token, designed specifically for agentic tasks such as tool calling and multi-step reaso…
Theory Ventures, a venture capital firm founded three years ago on the premise that AI would reshape software, reports that the market has exceeded bullish expectations, with frontier models moving to production, open-so…
A LangGraph pipeline can complete successfully while silently taking the wrong path, a failure mode standard monitoring fails to catch. The author argues that routing decisions should be logged as data, not just control …
FastMCP, a framework for building MCP servers, faces security and access control challenges when connecting to real company data. Developers must address authentication, logging, and tool selection to safely deploy MCPs …
OpenAI launched GPT-5.6 with three tiers (Sol, Terra, Luna) and ChatGPT Work, a unified professional agent, while Meta released Muse Spark 1.1 and Anthropic shifted to usage-based billing for Claude Fable 5. DeepSeek's r…
Nvidia is focusing on improving 'tokens per watt' as hyperscalers shift from building AI infrastructure to monetizing it, with offload engines designed to reduce GPU underutilization caused by memory bottlenecks. The met…
The state of AI networking in 2026, emphasizing that momentum in the field tends to persist. It explores trends and technologies shaping AI infrastructure and connectivity.
LangGraph consulting engagements follow a structured process: a discovery week to define objectives, an architecture sketch with risk assessment, a four-week prototype sprint, and optional productionization. The approach…
AI data centers are emitting illegal levels of air pollution in communities with little public oversight, as exemplified by xAI's Colossus 2 facility in Mississippi, which installed 33 gas turbines without required permi…
Tilion launches a stealth Chromium-based browser and Model Context Protocol (MCP) to prevent AI agents from being blocked by websites, aiming to give agents equal access to the web. The project argues that current web re…
A Hacker News user asks the community about practical local LLM usage, seeking real-world experiences with models, hardware, use cases, and trade-offs versus APIs.