Sources

Research log

The articles and reports that moved the atlas in the last 7 days — the methodology trail that keeps the project honest. 3 entries.

  1. 36KrQwen / Benchmarks
    Qwen's coding/agent story gets a benchmark-heavy update.

    36Kr frames Qwen3.7-Max as an agent-base-model push, with reported Code Arena rank, long-running tool-use, and low-cost coding-task evidence. The atlas treats this as source-attributed model context.

    #coding#agents#benchmarks
  2. 36KrByteDance Seed
    Doubao's phone episode sharpens the agent distribution question.

    ByteDance denied building its own phone, but 36Kr's report on the Nubia M153 Doubao AI device, app-access blocks, and mobile OS hiring makes system-level agent access a central Seed/Doubao question.

    #agents#devices#distribution
  3. 36KrQwen / ByteDance Seed
    Alibaba vs. ByteDance is less model-vs-model than cloud-vs-feedback-loop.

    36Kr reports Volcano Engine at 49.5% of China's enterprise MaaS token calls versus Alibaba Cloud at 28%, while also noting Alibaba's stronger infrastructure depth and customer trust.

    #MaaS#tokens#cloud