Key Info

Qoder switched the default reasoning effort for GLM-5.3, Kimi-K3, Kimi-K2.8-Preview and DeepSeek-Flash from Max to High, reporting a 40.5% lower cost per run and a 6.3% relative score drop versus Max on QoderBench, which covers nearly 200 cases.

Highlights

  • GLM-5.3 moved from Max to High: cost per run down 40.5%, score down 6.3% relative to Max on QoderBench (~200 cases).
  • Default is now High for GLM-5.3, Kimi-K3, Kimi-K2.8-Preview and DeepSeek-Flash.
  • GLM-5.3-Flash and DeepSeek-V4-Pro remain on Max.
  • For everyday work Qoder recommends leaving the setting on High; for hard problems, users can switch the model back to Max via the model picker or Model Management.