Memdoor: A team of 5 Qwen3.8-27B models beats Fable 5 on LiveCodeBench Hard (opens github.com)
The claim is that a team of five Qwen3.8-27B models orchestrated by Memdoor beats Fable 5 on LiveCodeBench Hard; it is the developer's own benchmark.
Context: DeepSeek says Harness reached 1.54 million daily users within three months of launch and processes about 92 trillion tokens a day, roughly 60 million per user.
From Bloomberg: DeepSeek says its open-source agent app Harness hit 1.54M DAUs on November 4, less than three months after launch, and uses ~92T tokens a day, or ~60M per user — DeepSeek, the Chinese AI disruptor, released striking adoption numbers for its AI agent infrastructure as it continues …
The claim is that a team of five Qwen3.8-27B models orchestrated by Memdoor beats Fable 5 on LiveCodeBench Hard; it is the developer's own benchmark.
YC startup Pomerado connects agents to any website without needing an API, and it is open source.
A Google Chrome team repo of guidance that steers coding agents toward modern browser primitives instead of 2019-era JavaScript.
Strands (AWS) says Amazon replaced Sonnet 4.6 with its Strands Decider in an agent workflow and saved millions; 689 points and 242 comments on HN.
Anthropic also announced Claude Whiteboards, Sheets and Data the same day; Claude Meetings drafts plans and timelines as participants talk.
Graphon builds memory for AI agents by mapping connections across docs, videos and images; RRE led.