Roseram NewsAll stories

Chinese & International · Qwen · Doubao · GLM

Qwen, Doubao and GLM are a three-lane race.

Alibaba’s Qwen, ByteDance’s Doubao and Zhipu’s GLM show up together on OpenCompass-style boards. 36Kr and Ken Huang map different slices of the same contest.

Tools: Qwen · Doubao · GLM

Three corporate logics

Qwen’s open-weight releases shape developer ecosystems. Doubao is entangled with ByteDance distribution. GLM’s Zhipu storyline mixes research branding with commercial API push. Treating them as interchangeable ‘China models’ flattens the market.

36Kr English pieces and Ken Huang’s agentic-reasoning notes are useful secondary maps—still subordinate to primary eval cards and vendor model cards.

How we write the race

Roseram lists the model family, release date, and whether a claim is about chat product, API weights, or benchmark-only checkpoints. Without that, ‘Qwen beat Doubao’ is incomplete.

Read the underlying record.

  1. 0136Kr English ↗
  2. 02Ken Huang — test-time compute ↗
  3. 03OpenCompass rankings coverage ↗