SYLIN

Qwen Flash

by Alibaba · Model · Language & code models

Qwen3.8 Flash described as 125B parameters with 6B active and a 1M-token context window

AccessOpen weightsSize125B parametersLicenseApache-2.0

Evidence

  1. Aug 29
    Recreated an image as a single HTML/CSS cats scene in about 32 minutes; more realistic than GLM, with no 3D attempt
  2. Aug 29
    Shares GLM 5.3 Flash's design: 3 of 4 attention layers linear, 4x context compression, 4 residual streams, Muon optimizer; keeps positional encoding

Alternatives

GLMFixed a logic bug in a Docker app via Hermes Agent, but took 15 minutes versus Qwen's 4 and tested fractional servings
More Qwen Flash