Anthropic publishes 11 open-source plugins that give Claude Cowork role-specific skills, slash commands, and connectors for jobs such as sales, finance, legal, and data. The Apache-2.0 repository pairs plugins with tools including Slack, HubSpot, and Snowflake, and users install them from claude.com/plugins. The company says teams can edit the markdown and JSON files to fit their own processes.
DeepSeek-AI published DeepInfer, the inference framework it says powers its production services, under Apache 2.0. It serves DeepSeek-V4.1-Flash on Nvidia SM100 GPUs and Huawei Ascend 950 NPUs. The release omits a KV cache system and prefill/decode disaggregation, which the repository says a complete service requires.
Google released TorchTPU, an open-source version of PyTorch that runs AI apps on its Tensor Processing Units with a configuration change and no rewriting. Amazon released similar software, TorchNeuron, for its Trainium chips a day earlier. Both companies want to loosen Nvidia's hold on AI development, where PyTorch is used by most AI apps.
Meta releases Muse Spark Open, an open-weight mixture-of-experts model with about 816 billion parameters, 42 billion active per token, and a 1 million token context window. It accepts text, images, audio and video and generates text, aimed at agentic coding. Meta calls it the most capable US open-weights model. The BF16 checkpoint is about 1.5 TB.
Ansub Khan releases tv-detox, an open-source home screen that turns off bloatware and ads on older Android TVs, with Claude Code or Codex running the setup. Since v1.4.1 the installer keeps Google TV's own home screen on by default, after a Chromecast got stuck restarting and needed a factory reset. Disabled apps are not deleted.
NVIDIA researchers release ASPIRE v1.0, an Apache-2.0 framework that treats robot learning as skill refinement rather than gradient descent. NVlabs reports 99% success on 80 LIBERO-Pro tasks and Rubik's Cube solving, with Claude Opus 5.5 reaching 97 to 100% across three benchmarks. It warns generated code runs without a security sandbox.
Gathin23's cleanyourtv disables bloatware on Android TV and Google TV over Wi-Fi and installs a native home-screen launcher styled after Apple TV. It needs no root and uninstalls nothing, so a restore script reverses each step. The launcher builds to about 250 KB, and the repo ships no prebuilt APKs.
HazyResearch publishes NCM, a two-stage text generation framework whose coarse-to-fine decoder predicts token hierarchies in parallel, with authors claiming 9-31x faster generation than strong diffusion baselines at matching quality. Code, checkpoints, and tokenized datasets are released under the MIT license on Hugging Face, requiring Python 3.10 or newer and a CUDA-capable PyTorch install.
Reflection releases Beam, a 501B-parameter open mixture-of-experts model that activates 23B parameters per token and handles 1M-token contexts under Apache 2.0. The company reports 81.6% on SWE-bench Verified at BF16 precision and offers $10,000 in Hugging Face Hub credits for the best projects built on it.
PostgreSQL 19 slips to about November 19 from late September, and developers have reverted eight to 10 features to version 20, including SQL/PGQ support. Committer Peter Eisentraut of EDB attributes the rework partly to LLM-assisted code review, which surfaced dozens of edge-case defects developers previously found only after release.