Anthropic published a misuse report on September 10 documenting seven categories of harm from abuse of its Claude models, including agentic AI-run cyberattacks, a romance scam operation using roughly …
Former Anthropic and OpenAI pretraining researcher Jacob Kahn resigned from Anthropic and posted on X on September 8 that both companies are racing toward self-improving superintelligence and gambling…
Anthropic engineers Barry Zhang and Mahesh Murag have moved away from building a separate custom agent for every use case, packaging reusable "agent skills" — folders of instructions, scripts, and ref…
Google DeepMind researchers released a public dataset mapping the Drosophila fruit fly brain, documenting roughly 166,000 neurons and more than 125 million connections, which hobbyists have since load…
Nanonets released Graft, an MIT-licensed open source CLI tool that builds a queryable code graph for AI coding agents, distributed as the @nanonets/graft package and requiring Node 20 or newer, with v…
Researchers trained a publicly released Drosophila connectome of roughly 166,000 neurons and more than 125 million connections to classify emails, reaching about 80% accuracy against a small conventio…
Testing on a Panther Lake mini PC showed that speculative decoding with a draft model more than doubled local LLM token generation in Llama.cpp, from about 5 tokens per second to roughly 11 on an inte…
Critics argue that "pacing the frontier" rhetoric from Anthropic and OpenAI leaders Dario Amodei and Sam Altman functions as regulatory capture that could eventually restrict open-weight and locally r…
Thesys published OUI-1, a 26B-parameter diffusion model with 4B active parameters finetuned via LoRA from Google's DiffusionGemma 26B-A4B-it, which generates UI screens in openui-lang and scores 71.7%…
Alvin Graylin, a technologist who has worked in AI on both sides of the Pacific since the early 1990s, argues the US-China AI arms race is largely a myth used to justify bad policy and overspending, p…
Anthropic lowered Claude Code's default reasoning effort from high to medium in March, fixed a caching bug that stripped older reasoning from long sessions, and reverted a response-shortening system i…
OpenAI's GPT-6 Astra requires noticeably less hand-holding than prior models like Opus, making fewer wrong assumptions about developer intent, according to hands-on testing described in a step-by-step…
Anthropic's Claude Opus 5 posted top scores on Anthropic's own benchmarks — more than doubling Opus 4.8 on its frontier coding test and tripling the next best model on a separate agentic benchmark — b…
A week of head-to-head testing found GPT-6 Astra outperforming Fable 5.1 on most real-world coding and reasoning tasks despite roughly equal benchmark scores, according to developer testers cited in t…
Token Rhythm's NeoHorse-1-4B, a 4 billion parameter model built on Qwen 3.5, was trained on live router decision logs rather than a static dataset, with a stronger teacher model correcting its attempt…
Edge0-35B-A3B, a 35B-parameter sparse mixture-of-experts model built on Qwen3.5-MoE, loses an average of 3.9 points across five benchmarks when quantized to 4-bit through its edge0 inference pipeline …
Nex-AGI released its Nex-N2.5 family of open-weight agentic models in three sizes — mini, Pro, and Max — with a prebuilt Docker image running a customized SGLang fork (nexagi/sglang:v0.5.18-nex-patch)…
Qwen released Qwen-Drive-1.0-4B, a 4-billion-parameter vision-language model built on Qwen3.5-4B that fuses perception and planning into a single model with a bird's-eye-view perception head and two p…
Cognition's SWE-2 coding model scored 83.75% on the eight-task KingBench 3 benchmark, beating DeepSeek V4.1 Flash at 81.25% and its own base model Kimi K3 at 77.5%, according to independent testing. S…
AI agents can safely customize existing Mac and Windows desktops when given scoped, limited control through tools with readable config files and documented commands, such as Aerospace on Mac, Apple Sh…