Ox Alpha

A model with a 1 million token context window, video input and mandatory reasoning appeared on OpenRouter on August 20 with no lab attached and no price. We pulled the platform metadata ourselves, matched its serving profile against every other model in the catalogue, and read the terms you accept when you send it your code.

DeepSeek Harness: The Open-Source Agent Layer Nobody Owns Yet

DeepSeek shipped the middle layer of the agent stack under MIT and told everyone to take it apart. Here is what dsh actually is, what the GitHub and npm APIs really say about its adoption, and why a free harness arrived three days before V4 output tokens got 4.5 times more expensive at peak.

MotionBricks

NVIDIA Research shipped MotionBricks, a 224M-parameter model that replaces the hand-wired animation state machine with one generative backbone: 350,000 motion skills at 15,000 FPS and 2 ms latency, driving both game characters and a Unitree G1. Apache 2.0 code, open weights, and a dataset catch worth knowing about.

DeepSeek-V4-Flash

DeepSeek-V4-Flash 0731 is official: MIT weights, 284B MoE (13B active), 1M context, and agent benchmarks above V4-Pro-Preview at $0.14/$0.28 per 1M tokens.

Kimi K3 Open Weights Are Live: What Actually Shipped

Moonshot put the Kimi K3 weights on Hugging Face, and the largest open-weight model anyone has shipped is now a download. We read the actual repo instead of the press cycle. Here is what landed, what it costs you in hardware, and which widely repeated numbers do not survive contact with the file listing.