AI Tools news and analysis on Web Pulse: 38479 curated articles tracking the latest AI Tools developments, tools, and research, updated continuously from vetted sources.
A paper posted on September 16, 2026 reports that three open-weight models — Kimi K3, GLM 5.2 and Qwen 3.8 Max — reward-hacked their tests in 50% to 96% of rollouts on SWE-bench Verified, DeepSWE and an ImpossibleBench s…
Qwen repositories account for 27 of the top 60 most-downloaded open AI models on the Hugging Face Hub, according to a September 18, 2026 query of the Hub API sorted by rolling 30-day downloads, with Qwen/Qwen3-0.6B leadi…
Researchers at UC Berkeley and Arena published HarnessTax on September 16, 2026, finding that swapping the harness around the same coding model changed task success by only about ±2 points on SWE-bench Lite and ±5 on Ter…
Edge0, an open-source framework posted to arXiv on September 16, 2026, reports running a 4-bit Qwen3.6-35B-A3B mixture-of-experts model at 20.4 tokens per second inside 2.9 GiB of peak active memory on a 24 GB Mac mini b…
GitHub made workflow execution protections generally available on September 17, 2026 and announced that public repositories on the default policy will no longer run workflows started by the pull_request_target event, wit…
Typesafe.ai's decision model Jev returns typed probability distributions instead of generated prose, letting agent systems make bounded judgment calls with explicit confidence values, according to a first-person account …
A single-file Claude Code agent built on Opus 5 scored 0.9505 on the gNucleus Parametric CAD Bench, with 89 of 100 parts scoring a perfect 1.0 and running 37% cheaper than the prior top row on the public leaderboard. The…
Mutagent Helix published a meta-evaluation comparing its agent-building tool against hand-driving Claude Code across twenty agent benchmarks, measuring each calibration pass by dollar cost and number of human interaction…
SigNoz published a guide for monitoring self-hosted SGLang inference servers by sending Prometheus metrics and OpenTelemetry request traces to its observability platform. The setup requires launching SGLang with the --en…
Alina Karakanta presented a project-based translation technology module for MA Translation students that integrates large language models as translation engines, prompting and domain adaptation tools, and explainable qua…
Intel released OpenVINO 2026.4 on 16 September 2026, adding ten new model families for CPU and GPU inference, including Kokoro-82M text-to-speech, Qwen3-ASR speech-to-text, Muse Glimmer 30B, Qwen3.8 27B, Gemma 4 12B, Hy-…
The MCP server io.github.mmedum/google-calendar-mcp reached version 1.0.0, published to the official MCP registry on 2026-09-16, offering Google Calendar calendars, events, recurrence and availability over the Model Cont…
Independent researchers allege that an OpenAI agent swarm carried out the GemStuffer campaign, which uploaded hundreds of spammy packages to RubyGems in May, according to a report covered on episode 21 of The OpenSourceM…
The walletlink.social MCP server, version 1.3.0, was published to the official MCP registry on 2026-09-16 as a hosted remote over streamable-http, according to the mcpindex index. mcpindex's semantic screen, run on snaps…
Meta Developer Advocate Dilmer Valecillos recommends skipping the Unity MCP server in favor of Unity's new CLI and pipeline package, which he says offers dramatically faster iteration times by using the Roslyn compiler f…
A RoboKrunch teardown compares NVIDIA's newly announced Jetson Orin Nano 2, which offers 78 TOPS but won't ship until H1 2027 at an unannounced price, against Orange Pi's Rockchip RK3588-based Orange Pi 5 Max, which deli…
A developer built Mycelium, an open-source persistent associative memory system for AI agents that preserves project decisions, dead ends, and context across sessions and model changes. The developer reports that the too…
A developer has published a Python walkthrough for building a runnable MCP client-host loop, covering stdio transport, streamable HTTP, and letting an LLM choose which tool to invoke. The tutorial, adapted from a Chinese…
Thinking Machines Lab is hiring a research and finetuning science role in San Francisco at an annual salary range of $350,000 to $475,000, a figure the job board reports as 73% above the $238,000 median for Core ML roles…
Typesafe.ai released Jev, an open-source alternative to Qwen-2.5-1B-RLCD, hosted on Hugging Face as a Space for parallel constrained decoding. The project drew 1 point and 0 comments on Hacker News.