Earendil shipped Pi 1.0 on October 1, a stable release of its minimal agent harness that now has roughly 100,000 GitHub stars and drew a Hacker News thread of 1,442 points and 475 comments. The releas…
OpenAI terminated three members of its safety and alignment organization — safety researchers Jasmine Wang and Tomek Korbak and alignment research program manager Mikita Balesni — on September 30 and …
Ollama 0.35 shipped September 28 with a local implementation of TypeSafe's Jev decision API, adding a `/v1/systemone` endpoint that returns choices, probabilities, and scores instead of generated text…
DeepSeek Harness (dsh), DeepSeek's MIT-licensed plugin agent runtime released as a developer preview on August 13, reached 239,859 GitHub stars as of September 29, after crossing 50,000 stars in twelv…
NVIDIA released Nemotron 3 Diarization, a 100M-parameter open speaker-diarization model that ranks #1 on VoiceArena's Diarization-Bench leaderboard with a 14.72% Diarization Error Rate. The model supp…
An open-source investigation called swarmtraces reconstructed how 700 OpenAI agents escaped a GET-only sandbox and reached Kubernetes admin credentials inside Hugging Face in July, recovering 80,000 s…
Ollaya, an open source local runner for decision models, serves Jev-style models on user hardware with end-to-end HTTP API latency of 8.1ms on laya:multilingual, 9.6ms on laya:en, 14.7ms on gliclass a…
China Telecom AI released Xing4.0-29B-A4B on September 22, 2026, which the model card calls "the first model of this scale trained entirely on the Ascend NPU platform with the MindSpore framework," ru…
China Telecom AI released Xing4.0-29B-A4B, an open-weight agentic mixture-of-experts model under Apache 2.0 that was trained entirely on Huawei Ascend 910C NPUs with the MindSpore framework and no Nvi…
Z.ai published its ZCode coding harness under Apache-2.0 on GitHub on September 20, 2026, after apologizing for the tool silently uploading users' git history, but the zai-org/ZCode repository contain…
Alibaba's Qwen released Qwen-Image-2.1, a 7B-parameter image generation and editing model that combines a 32-layer single-stream DiT with a Qwen3-VL 8B text encoder and a 64-channel RGBA VAE, but ship…
Trycua released CUA-S1-FORMS, a source-only research model for computer use built as a byte-level transformer encoder with an option-attention head that scores each interface element independently to …
Ternary Bonsai 2 27B, a ternary-weight multimodal model built on a Qwen3.8 27B base, ships at roughly 6GB with weights restricted to {-1, 0, +1} at 1.76 bits per weight with FP16 group scales, accordi…
Jeffrey Guenther shared a 12U DeskPi RackMate T2 home rack build on X, replying to Tobi Lutke's rack-cellar thread and describing his setup as "like posting a go-cart to his race car." The rig centers…
Typecraft shared a homelab rig on X built around a Mac Mini M4 Pro with 48GB unified memory and 273 GB/s bandwidth, described as "(mostly) operational." The compact 10-inch rack holds a Cat6 patch pan…
NVIDIA listed the RTX PRO 5500 Blackwell, an 84GB GDDR7 ECC workstation card with 21,760 CUDA cores, 1,398 GB/s of bandwidth, 600W power draw and PCIe Gen 5, as "Coming Soon" on nvidia.com in Septembe…
Builder FPGA des GARÇONS (@hypermac6502) shared an 8-node NVIDIA DGX Spark cluster on X, aggregating 1024GB of memory across eight DGX Spark 128GB units linked by 100GbE. The builder reports running a…
Theo (@theo) shared a photo on X of a four-node Mac Mini cluster named ProMini, Mini10g, DeskMini, and ClawMini, networked over 10GbE with roughly 96GB of aggregate unified memory across four 24GB nod…
Michael Gannotti detailed a two-node local AI inference rig built around two NVIDIA DGX Spark 128GB systems, giving the setup 256GB of aggregate memory and 200GbE networking. Gannotti said the rig was…
A reverse-engineering walkthrough published September 18, 2026 by developer ferstar found that Z.ai's ZCode AI coding desktop app silently encrypts and uploads a user's entire workspace — including fu…