Light Origins shipped Light-O1 on Sept 21: pretrain on recovered human actions from internet video, then adapt the same base to LightBot and Unitree G1. It also released Light-O1-Preview (6B, Apache 2.0) on Hugging Face for text-to-motion; full robot demos still need separate controllers.
Cloudflare made Python Workers generally available on Sept 21, so FastAPI, Django, and Flask can talk to R2, D1, Workers AI, Queues, and the Python MCP package without a JavaScript conversion layer. Native C/C++/Rust packages still need PyEmscripten wheels…
RuntimeWire reverse engineered Codex desktop build 9922 and found a feature gated Cloud path (vs Cloud Legacy) that lets the agent draft environments, then attach Tailscale auth keys, proxy delivered network secrets, and Azure OIDC workload identity. Public docs still describe manual legacy setup; launch timing is unconfirmed…
Hacktron researchers used Claude Opus 5 to help chain a Discourse/libheif image bug into OpenAI staff ChatGPT and Codex accounts, then opened a harmless demo PR in an internal repo. OpenAI patched within about 14 hours and paid a $6,500 bounty on Sept 1; there is no sign the chain was abused in the wild…
Ethan+ released an open source Codex skill that keeps GPT 6 Astra on planning and review while DeepSeek V4.1 Flash handles implementation and tests. In one local build he measured about a 98.9% drop in Astra input tokens per 1,000 implementation lines—a single uneven field comparison, not a controlled bench…
Yandex published AliceAI Foundation 80B A3B Base on Hugging Face: an 80B MoE (3B active) pretrained from scratch with 262K context, Apache 2.0 weights, and Transformers/vLLM paths. It is a base model only—no post training—and ships Russian factual benchmarks WikiWebFacts and HardMultiQA…
Spotted something we missed? Start a thread.