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…
David Heinemeier Hansson, creator of Ruby on Rails and co-owner of 37signals, released Omarchy, a free and open-source Arch Linux-based distribution that ships ten AI coding-agent command-line tools p…
Tencent released Hunyuan Hi-4 Preview, a 770B-parameter mixture-of-experts model with 49B active parameters and a 1M-token context window, under the Apache 2.0 license. The model scores 92.3 on GPQA D…
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 techniqu…
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 p…
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 i…
Omarchy, a Linux distribution, now routes crash dumps from systemd-coredump to a user's default coding agent via a desktop notification, attaching a 'diagnose-crash' skill to guide analysis. Users can…
Omarchy 4, a Linux distribution, allows users to set Claude Code as the default coding agent with the command `omarchy default agent claude`, or via Setup › Defaults › Agent in the Omarchy Menu. The s…
OpenAI, Nvidia, and Anthropic are pursuing distinct AI compute strategies: OpenAI revealed its custom inference chip Habanero, claiming it beats Nvidia's GB200 and GB300 systems on latency and through…
An AI software factory, or 'dark factory,' is a pipeline that converts a product requirements document into deployed code without human review, according to a concept popularized by Dan Shapiro's five…
OpenAI, Nvidia, and Anthropic are pursuing divergent strategies in AI infrastructure, with OpenAI vertically integrating by developing its own inference chip (code-named Habanero) and cutting off futu…
Anthropic's Claude Opus 5.1, released September 1st, scored 92.5% on an independent coding benchmark, the highest recorded on that test, surpassing GLM 5.3 (91.1%), Opus 5 (82.5%), and Qwen 3 Max (81.…
Anthropic's Claude Opus 5.1 model is being used by developers to generate playable browser games and 3D scenes in a single prompt, including a class-based FPS shooter by Riley Brown, a Mario Kart clon…
Anthropic's Claude Opus 5.1 supports six reasoning effort levels (low, medium, high, extra, max, ultra) that developers can switch between mid-session using Control+Shift+E, with hands-on testing show…
Google's Gemini 3.8 Flash, released three weeks after Gemini 3.7 Flash, scores 73.7% on Deep SWE v1.1, edging out GPT 5.6 Soul's 72.7% and matching Claude Opus 5, while priced at $0.75 per million inp…
Google's Gemini 3.8 Flash, released three weeks after Gemini 3.7 Flash, scored 73.7% on the Deep SWE coding benchmark, roughly matching Claude Opus 5's 74% and ahead of GPT 5.6 Sonnet's 72.7%, at a co…
Google's Gemini 3.8 Flash, priced at roughly 75 cents per million input tokens, matched or beat pricier frontier models from OpenAI and Anthropic in hands-on tests covering agentic debugging, vision-b…
Meta's Muse Spark 1.3, priced at $1.25 per million input tokens and $4.25 per million output tokens, outperformed GPT-5.6 and Opus 5 on a million-token long-context retrieval test, scoring in the high…
Claude Opus 5.1 can generate premium-feeling websites with layered scroll animations and dynamic text effects, but only when given concrete design references and reusable skill files, according to a p…
Anthropic's Claude Opus 5.1 model shows significant benchmark gains over its predecessor Opus 5, with scientific research capability roughly doubling to over 50% on Terminal Bench Science 1.0 and busi…