TRACER, a recent arXiv paper, takes the input/output logs of an LLM classification endpoint and reuses them as training data, then swaps in a lightweight surrogate only on regions that pass a parity gate to cut inference cost. The surrogate absorbs 83–100% of traffic on a 77-class intent dataset and 100% on a 150-class one, while correctly refusing to deploy on an NLI task — that refusal behavior is the interesting part.
Japan's Digital Agency released parts of Gennai, the generative AI platform it runs for central-government staff, on GitHub under MIT / CC BY 4.0. The web app and cloud-specific AI templates for AWS, Azure, and Google Cloud are bundled together so local governments and private companies can redeploy the same stack.
DeepSeek V4 Preview ships V4-Pro (1.6T/49B active) and V4-Flash (284B/13B active) as open weights under MIT, both with 1M context. CSA+HCA hybrid attention, mHC, and the Muon optimizer cut per-token FLOPs at 1M tokens to 27% of V3.2. Day-one API and chat.deepseek.com mode switch covered.
Two open-weight Chinese MoEs landed within 24 hours: Ant Ling-2.6-flash (104B/7.4B active, 7x token-efficiency claim) and Tencent Hy3-preview (295B/21B active, frontier-tier open weights). Specs, licenses, and how they line up against DeepSeek-V3 and GLM-4.5.
Xiaomi launched two MiMo-V2.5 models at once. MiMo-V2.5-Pro hits SWE-bench Pro 57.2, Claw-Eval 63.8, and τ3-Bench 72.9 — frontier-tier — while MiMo-V2.5 brings native omnimodality plus a 1M context. Both are API-only for now; open weights are promised but unscheduled.
NVIDIA's build.nvidia.com serves a free inference API that covers 100+ models including MiniMax M2.7, GLM-5, Kimi K2.5, DeepSeek, GPT-OSS, and Sarvam-M. Because integrate.api.nvidia.com/v1 is OpenAI-compatible, OpenClaw, OpenCode, Zed, and Cursor can call it directly.
The NotebookLM clone open-notebook assumes Docker and cloud APIs by default. I installed SurrealDB natively, ran four processes in tmux, and wired everything through Ollama's qwen3.6:35b and bge-m3. I fed it the Qwen3.6 benchmark article I wrote this morning, and it answered with the correct numbers.
A dev.to post about building a Moon Patrol-style 80s game in a few days with AI for an internal AWS Summit contest. The Phaser 3 + TypeScript + Vite stack, combined with splitting into three role-specific AI skills instead of one giant prompt, turned out to be a practical workflow.
SwitchBot released an official CLI for OpenAPI v1.1. It does device listing, control, and scene execution from the terminal, and also runs as an MCP server so Claude or Cursor can drive home devices directly.
A walkthrough of NeuroValkey Agents, a multi-agent swarm wired up with OpenAI API, Valkey, and Node.js. Not a cache — Valkey here is the orchestration, memory, and state-transition substrate for the whole system.
Tried Qwen3.6-27B on both Ollama and MLX. Ollama couldn't load the VL-projector-embedded GGUF, MLX ran it at 11 tok/s. On the side, running 35B-A3B under MLX was roughly 2× faster than the Ollama GGUF. Also had both models build a BBS to gauge intent handling.
A hub for the 5-article series that organizes math symbols in AI and LLM articles for reading, not solving. Covers equations, vectors and matrices, probability and statistics, derivatives, and gradient descent with backprop, plus a reading-order guide for different backgrounds.
