A few months ago, I started helping my parents organize their health information. As they get older, I realized how much medical history builds up over time: test results, medication changes, doctor visits, and small details that seem insignificant when they happen but become important later. The frustrating part was not that we lacked information. We actually had too much. Some records were in hospital portals, some were saved as PDFs, some were buried in emails, and some details only existed i...
We just released the first public results from the Agent Memory Leaderboard (AML). The first evaluation focuses on Text Memory across two tracks: - Open-source Methods
- Commercial Products 136 teams registered, and 69 representative memory frameworks completed the first evaluation. Commercial Products β Text Memory: 1. MemoraX β 58.02
2. MemOS β 45.89
3. NTES-MEMORY-SMART β 44.21 The benchmark uses a common evaluation framework and a clearer system boundary: Memory system: Add β Search
Benchmar...
The model releases this week looked like one race on the surface, but the business models are separating. xAI's Grok 4.6 is a closed API product. Its leverage comes from controlling access and setting the token price. Alibaba's Qwen3.8 is an open-weight release. Its leverage comes from adoption by organizations that want to operate or adapt the model themselves. Nvidia's Switchyard is a routing layer. It sends each job to the cheapest model that can handle it, which means Nvidia can influe...