Z.ai
Z.ai: GLM 4.5 Air
z-ai/glm-4.5-air
GLM-4.5-Air is the lightweight variant of our latest flagship model family, also purpose-built for agent-centric applications.
- Context window: 131,072 tokens
- Input: text
- Output: text
- Pricing: $0.13/M input tokens, $0.85/M output tokens
Community sentiment
Positive (+57) — 3 positive, 1 negative, 0 neutral across 4 community samples from X and the web, as of 2026-10-02.
Discussion highlights strong post-training gains and community recipes for GLM-4.5-Air (notably via Rufus-Air and an open-sourced Amazon recipe that boost benchmark scores), while some users flag slower decoding throughput compared with alternatives. Availability on platforms and SOTA claims for reasoning and code are also cited positively.
- “Amazon open-sourced a full post-training recipe for GLM-4.5-Air and it wrecks the official release. Rufus-Air: 8 stages from SFT through agent RL to RLHF. Public data only, no new human labels. Same 106B/12B MoE base. IFBench jumps from 33.6 to 76.9. SWE-bench Verified from 50.6” — X
- “Rufus-Air is a blueprint for building strong open LLMs with nothing but public data and accessible compute—no secret sauce, no new human labels, no in-house distillation. The team takes a 106B MoE base (GLM-4.5-Air-Base) and walks it through 8 post-training stages, from massive s” — X
- “Blogs - SiliconFlow | AI Insights and Tutorials for Developers — 5 days ago · GLM-4.5 Now Available on SiliconFlow: Open-Source SOTA Model for Reasoning, Code, GLM-4.5-Air, Z.ai's latest flagship model serie, into SiliconFlow platform.” — the web
- “gpt-oss-20b ran 8,287 tokens per second on openllms B5 prefill across our RTX PRO 6000 roster. glm-4.5-air measured 1,408 under the same card family. That spread sits next to the decode column on https://t.co/FuD3490vd3. https://t.co/IgfvHaC9PZ” — X
View on OpenRouter.
Model data sourced from OpenRouter.