AI Infrastructure news and analysis on Web Pulse: 35084 curated articles tracking the latest AI Infrastructure developments, tools, and research, updated continuously from vetted sources.
Netlify added Anthropic's Claude Fable 5.1 and Google's Gemini 3.8 Flash to its AI Gateway and Agent Runners, but independent comparisons still favor Vercel's broader gateway, which offers dozens of models, bring-your-ow…
A developer running My Mind is Racing, which tracks over 53,000 events and 22,000 organizations, details a cost-cutting approach to AI-assisted web scraping that minimizes LLM usage. The method prioritizes JSON-LD data, …
AMD and the veRL project released a production-ready reinforcement learning post-training container for AMD Instinct GPUs, supporting MI300 and MI355 series on ROCm, with AITER-accelerated vLLM and SGLang rollout backend…
Elastic NV announced the Elastic{ON} 2026 tour, with events in Mumbai, New York, Amsterdam, San Francisco, London, Washington D.C., Singapore, and a digital finale from September 2025 through March 2027. CEO Ash Kulkarni…
OpenAI's research staff logged 3.14 agent-workdays for every 8-hour human shift in mid-August, with the median researcher's daily inference cost surging from $14 in late March to over $600 by mid-August—a 40-fold increas…
ModelBench's MiniCPM5-2B, a 2-billion-parameter dense language model from the Tsinghua University spinout, can run locally with SGLang or llama.cpp, requiring roughly 4GB of VRAM for weights but up to 44GB with default K…
Das Lab at the Institute of Science and Technology Austria (ISTA), the team behind GPTQ, introduced GSQ (Gumbel Softmax Quantization) and RCO (Riemannian Constrained Optimization), two linked techniques that compress lar…
WorkBuddy, a desktop AI agent app, is offering free access to Tencent's Hi-4 Preview model for two weeks from launch, allowing users to try the 770 billion parameter mixture-of-experts model without payment. The model, r…
Ray Summit 2026, held August 24-26 at the San Francisco Marriott Marquis, drew more than 2,000 attendees to focus on reinforcement learning as a production engineering problem, with Robert Nishihara, co-creator of Ray an…
The Omacom Foundation, the nonprofit behind DHH's Omarchy Linux, has raised roughly $15.5 million in total pledges as of early September 2026, including about $13.5 million in cash and $1.95 million in AI model tokens. F…
Meta Platforms Ireland's AI training crawler, meta-externalagent/1.1, overloaded the D1 database of Usero, a feedback tool built on Cloudflare Workers, causing 4,227 D1_ERROR events over 14 days and consuming 81% of the …
OpenRouter's monitoring of the deepseek/deepseek-v4-flash-0731 endpoint at mancer/fp8 recorded a B3IT level shift of 0.36 total-variation distance, while the LT logprob probe failed because the endpoint returned no usabl…
CIQ shipped Fuzzball 4.2 on September 3, adding a native Model Context Protocol (MCP) server that lets AI agents inspect, submit, and monitor HPC jobs directly, plus a self-service workflow API enabling jobs to spawn oth…
A developer has turned a Steam Deck into a server node for the BarkVisor homelab platform, enabling it to run VMs and local AI models via Ollama while remaining a gaming device. The walkthrough details installing BarkVis…
Lenovo has introduced the Yoga Pro 9n laptop, which can run a 120-billion parameter AI model locally with a 1-million token context window, powered by NVIDIA's RTX Spark chip featuring a 20-core Grace CPU and 6,144 CUDA …
A developer's blog post argues that AI security failures often stem from poorly managed data flow rather than model flaws, and outlines risks at each stage from ingestion to inference. The post emphasizes that treating d…
OpenCode v2 beta, a complete rewrite by Anomaly, replaces Bun with Node.js and Tauri with Electron to fix memory bloat and rendering inconsistencies, and introduces a service-first architecture with multi-tab parallel se…
At IFA 2026, AMD unveiled new AI-focused hardware including Strix Halo and Gorgon Halo chips, the Threadripper Halo Station with 96 cores, and announced Project Zenith with Microsoft's Pavan Davuluri, enabling local exec…
A developer built TraceShrink, an OpenTelemetry Collector processor that replaces random tail sampling with cost-aware rules to cut LLM observability costs. The processor keeps traces based on cost thresholds, errors, an…
Supermemory launched Learner-1 on September 7th, a continual-learning layer that lets AI agents carry experience across sessions, extending its memory infrastructure beyond retrieval. Founder Dhravya Shah announced the s…