Pre A

DeepSeek open-sources Harness agent runtime with plugin design: DeepSeek has released DeepSeek Harness, an MIT-licensed agent runtime currently in developer preview. The system features an everything-is-a-plugin design where models, tools, sessions, sandboxes, and the UI are Cordis plugins, with npm packages advancing through the v0.1.5-rc line.

AI Models and Applications • Pandaily

Researchers introduce ZGCM-1 open 7B model for math and search: Researchers have presented ZGCM-1, a fully open 7B dense foundation model trained from scratch for math and agentic search. The model overcomes parametric capacity limits by coupling internal thinking with active external tool use across a 256K context window.

AI Models and Applications • arXiv

Nvidia Vera Rubin NVL72 beats Blackwell by 7x in DeepSeek inference test: Inference tests on a 1.6-trillion parameter DeepSeek model show Nvidia's Vera Rubin NVL72 delivers up to seven times better token throughput per megawatt compared to Blackwell. This performance exceeds CEO Jensen Huang's previous claim of a threefold improvement for models ranging from 1T to 3T parameters.

Hardware • Techmeme

Managed Agents - Don't Get Locked In: This video analyzes the concept of managed agents currently being rolled out by various Large Language Model providers. While these managed services offer numerous advantages, they also present a significant risk of vendor lock-in for developers.

LLM Agents • Sam Witteveen

Shanghai AI Lab and SJTU open-source 8.9B NCP-ArchPreview model: Shanghai Artificial Intelligence Laboratory and Shanghai Jiao Tong University's LUMIA Lab have open-sourced NCP-ArchPreview. This 8.9B latent-space model jointly trains next-concept and next-token prediction. It successfully matches a 7B baseline loss on approximately 51.3% of tokens.

open source • Pandaily

Study links AI written reasoning steps to distinct internal patterns: A new study reveals that reasoning steps such as calculation, formula retrieval, and deduction are clearly separable within an AI model's internal states, particularly in its middle layers.

Artificial Intelligence • The Decoder

Anthropic • Nvidia

Nvidia plans up to $10 billion investment in Anthropic's IPO: Nvidia is in talks to invest up to $10 billion in Anthropic's planned initial public offering, according to a report from Reuters. The AI startup is reportedly targeting a valuation of $2 trillion for the debut. If realized, the transaction would represent the largest IPO in history.