Open Source

Zhipu AI may skip GLM 5.2 Air, focus on 500B+ full and 30B flash models

Leaked Discord chat suggests a strategic pivot: Air model shelved for now.

Deep Dive

According to an unofficial conversation on Z.ai's official Discord, the company appears focused on full-size (500B+) and flash-sized (~30B) models, with the turbo model reportedly closer in parameters to flash than to Air.

Key Points
  • Zhipu AI deprioritizing the GLM 5.2 Air series, focusing instead on a ~500B+ full model and a ~30B flash model.
  • The turbo model is reportedly closer to flash size (~30B) than to the previous Air model size (~7-13B).
  • Strategy mirrors industry trends: huge models for research and benchmarks, small models for real-world deployment at scale.

Why It Matters

Signals Zhipu AI's bet on two extremes: massive flagship models and efficient, deployable flash models.

📬 Get the top 10 AI stories daily