z.ai2025-07-28officialauthorityHN front pageabout the GLM 4.5 familyHN 247 pts · 134 comments5,014 words
Z.ai introduces GLM-4.5 and GLM-4.5-Air, hybrid reasoning models designed to unify reasoning, coding, and agentic capabilities. GLM-4.5 features 355 billion total parameters, 32 billion active parameters, and 128k context length with native function calling capacity. The model achieves a 90.6% average tool calling success rate and a 53.9% win rate against Kimi K2 across 52 coding tasks.written, reviewed, cited
- GLM-4.5 evaluation for SWE-bench Verified uses OpenHands v0.34.0 with runs limited to 100 iterations.
- The HLE benchmark evaluation for GLM-4.5 only assessed text-based questions judged by gpt-4o.
docs.z.ai2025-09-06officialauthorityHN front pageabout the GLM 4.5 familyHN 213 pts · 84 comments1,048 words
Z.AI released GLM-4.5 and GLM-4.5-Air, foundational models optimized for agent-oriented applications using a Mixture-of-Experts architecture. GLM-4.5 features 355B total parameters with 32B active, while GLM-4.5-Air uses 106B total and 12B active parameters. Both support a 128k token context length, hybrid reasoning modes, and API costs as low as $0.2 per million input tokens and $1.1 per million output tokens with generation speeds exceeding 100 tokens per second. The series ranks second globally among all models, first among domestic models, and first among open-source models.written, reviewed, cited
- GLM-4.5 achieves higher parameter efficiency than DeepSeek-R1 and Kimi-K2 despite having half and one-third their respective parameter counts.
- The GLM-4.5 series lies on the Pareto frontier for performance-to-parameter ratio, delivering optimal performance at the same scale.
- GLM-4.5-Air surpasses Gemini 2.5 Flash, Qwen3-235B, and Claude 4 Opus on reasoning benchmarks like Artificial Analysis.
