Open-Source AI Models Advance with GLM-5.2 and Moebius Releases

오픈소스 | Tue Jun 23 2026 00:00:00 GMT+0000 (Coordinated Universal Time) | 2 sources

Z.ai released the 744B parameter GLM-5.2 with local execution support, while a 0.22B lightweight inpainting model Moebius was unveiled.

Analysis

[Z.ai / Unsloth] released GLM-5.2 open source model with local execution support ^[1]

• 744B total parameters
• 40B active parameters
• 1M context window
• Benchmark performance on par with Claude 4.8 Opus
• GPT-5.5
• and Gemini 3.1 Pro
• Unsloth Dynamic GGUF compresses 1.51TB down to 239GB (2-bit)
• Supports three thinking modes: Non-thinking
• High
• and Max

[Moebius (HUST & VIVO AI Lab)] released a 0.2B lightweight image inpainting open source framework ^[2]

• Matches or exceeds the 11.9B FLUX.1-Fill-Dev with only 0.22B parameters
• Introduces Local-λ Mix Interaction (LλMI) block
• 26.01ms/step
• accelerating overall inference speed by more than 15x
• Applies adaptive multi-granularity distillation strategy

Sources

• [1] Unsloth GLM-5.2 – How to Run Locally - Hacker News
• [2] Moebius: 0.2B image inpainting model with 10B-level performance - Hacker News