The extension has grown from 15M users in September to 22M, surpassing AdGuard, 360, and NordVPN.
The collection includes Base, instruction-tuned, and vision-language versions for both the 15B and 2B model sizes, with GGUF files available.
Training 200+ models from scratch shows that vision-only parameters are essential for VQA performance, while separate encoder pretraining provides no benefit at matched compute.
The update adds multi-model routing for DeepSeek V4.1 Flash and video support alongside the new dense model.
Marimba is a fast, lightweight speech model that returns typed decisions using a frozen Qwen3-0.6B backbone.
The model runs in-browser with response times in the tens of milliseconds on iPhone and outperforms larger competitors.
ByteDance's Coze Enterprise Edition will stop accepting new customers on Nov. 30 and terminate existing packages by Dec. 31, 2027, amid a broader industry shakeout.
The family includes a 9.5B Lite variant and supports ComfyUI, Diffusers, and MLX runtimes.
The 4B parameter model processes audio and text in 133 ms frames using a single sampling step for continuous dialogue.
The Apache-2.0 tool creates copy-on-write branches for MySQL and PostgreSQL in seconds, supporting AI agent sandboxes and CI workflows.
The update improves lip-audio synchronization and reduces error accumulation to support coherent video generation over 10 minutes.
AWS Emulator supports integration testing by mimicking AWS APIs locally without requiring credentials.
Gair NLP announces 801-30B-A3B-Base, a 30B-A3B MoE model trained from scratch on 11.8T tokens using recursive self-improvement (RSI) in pre-training, which is set to be open-sourced soon.
Successfully simulated can transport using the UR17G and Robotiq gripper in Isaac Sim 6.2.
The tool separates proposal generation from state acceptance using independent verification and Rust-native governance.
Vercel's November AI Gateway report shows decision models processed 26% of requests while accounting for only 0.53% of total spending.
devvm 0.1.0 creates per-project Lima VMs to run coding agents in isolation while securely storing tokens on the host.
New models offer $8/$40 per million token pricing and outperform Opus 5.5 on FrontierCode 2.0, while Claude gains meeting participation and spreadsheet capabilities.
OpenAI has launched business ownership registration and agent communication features in ChatGPT as a beta.
AI Briefing