GLM-4.5-Air
- In Input modalities: Text Out Output modalities: Text
- Reasoning
- Reasoning can be off
- Tool use
- Open weights
Prices
3 offerings · prices as listedglm-4.5-air| Conditions/ 1M tokens | Input | Cache read | Output |
|---|---|---|---|
| Standard | $0.20 | $0.030 | $1.10 |
Prices from Pricing (opens in a new tab) (checked Nov 1, 2026)
glm-4.5-air| Conditions/ 1M tokens | Input | Cache read | Output |
|---|---|---|---|
| Prompts over 32K tokens | ¥1.20 | ¥0.24 | ¥8.00 |
Prices from 智谱 BigModel：API 定价 (opens in a new tab) (checked Nov 2, 2026)
Standard: the price a request with no conditions pays (the headline price elsewhere on the site).
Details
- Knowledge cutoff
- Apr 2025
- Family
- glm-air
- Model ID
Not stated on Zhipu AI’s pages: max input, reasoning efforts, max thinking budget, structured output, built-in tools, licence, aliases.
Run locally
Weights: zai-org/GLM-4.5-Air- Parameters 110.5B
- Active 17B (estimated) · 128 experts, 8 per token
- Native precision BF16
- Layers 46
- KV cache per token 184 KB
- Config positions 128K (config.json max_position_embeddings)
- API context 125K (maker’s API)
- Needs
- ≈ 58.8 GB (weights 54.7 + KV cache 1.4 + overhead 2.7)
- Usable
- 21.6 GB (90% of its memory)
- In system memory
- 49.3 GB · the MoE experts in system memory, attention and shared layers on the GPU (llama.cpp --cpu-moe)
- Generation speed
- 12–18 tok/s (reads the active 17B per token)
- First token
- about 0.55–0.83 s for a 2K prompt
| Precision | Weights | Needs at 8K | On your hardware | Longest context |
|---|---|---|---|---|
| BF16 (native)maker’s file | 205.8 GB | ≈ 217.5 | ✕Doesn’t fit | — |
| Q8_0 | 109.3 GB | ≈ 116.2 | ✕Doesn’t fit | — |
| FP8maker’s file | 104.8 GB | ≈ 111.5 | ✕Doesn’t fit | — |
| Q6_K | 84.5 GB | ≈ 90.2 | ✕Doesn’t fit | — |
| Q5_K_M | 73.2 GB | ≈ 78.3 | ○Needs system memory 5–7 tok/s | — |
| Q5_0 | 70.7 GB | ≈ 75.7 | ○Needs system memory 5–7 tok/s | — |
| Q4_K_M | 62.5 GB | ≈ 67.1 | ○Needs system memory 10–15 tok/s | — |
| Q4_0 | 57.9 GB | ≈ 62.2 | ○Needs system memory 11–17 tok/s | — |
| IQ4_XS | 54.7 GB | ≈ 58.8 | ○Needs system memory 12–18 tok/s | — |
| Q3_K_Mquality drops | 50.3 GB | ≈ 54.2 | ○Needs system memory 13–19 tok/s | — |
| IQ3_XXSquality drops | 39.4 GB | ≈ 42.8 | ○Needs system memory 16–24 tok/s | — |
| Q2_Kquality drops | 38.6 GB | ≈ 41.9 | ○Needs system memory 17–25 tok/s | — |
| IQ2_XXSquality drops | 26.5 GB | ≈ 29.3 | ○Needs system memory 28–42 tok/s | — |
Click a row to choose a precision. “Maker’s file” sizes are the maker’s own files; others are parameters × bits per weight.
Compared with other hardware
Each at the highest precision it runs. Click one for details.
Smallest that runs it (Q4_K_M, 8K): GPU NVIDIA H100 SXM; unified memory Apple M5 Ultra (64-core GPU) · 96 GB.
Run commands
The maker publishes no files at this precision. Choose its native precision or one of its own builds, or quantise it yourself with llama.cpp.
Memory to fine-tune
Batch 1, 2,048-token sequences, gradient checkpointing; LoRA and QLoRA train rank-16 adapters on the attention projections.
Estimates; facts from the maker’s Hugging Face repositories, as of 2026-11-06. How it’s estimated
Cited scores
From third-party leaderboards, not measured by us- LMArena Text 1,373
#117 of 163 listed models · CI 1,369–1,377 · 30,652 votes
- LMArena Text · Chinese 1,405
#110 of 161 listed models · CI 1,393–1,418 · 2,234 votes
- LMArena Text · Coding 1,427
#116 of 163 listed models · CI 1,419–1,434 · 6,099 votes
- LMArena Text · Creative writing 1,327
#118 of 163 listed models · CI 1,318–1,337 · 3,996 votes
- LMArena (Arena), leaderboard-dataset on Hugging Face, licensed CC BY 4.0 · License · as of Oct 28, 2026
- changes we made
Changes
All changes →No changes recorded yet.
Sources
- Pricing (opens in a new tab) docs.z.ai · checked Nov 1, 2026
- 硅基流动：模型价格in Chinese (opens in a new tab) siliconflow.cn · checked Nov 1, 2026
- 智谱 BigModel：API 定价in Chinese (opens in a new tab) docs.bigmodel.cn · checked Nov 2, 2026
- 智谱 BigModel：模型概览in Chinese (opens in a new tab) docs.bigmodel.cn · checked Nov 4, 2026
- Z.ai LLM GLM-4.5 overview (opens in a new tab) docs.z.ai · checked Nov 1, 2026
- models.dev: glm-4.5-air (community data, MIT) (opens in a new tab) github.com · checked Sep 30, 2026
- GLM-4.5: Reasoning, Coding, and Agentic Abililties (opens in a new tab) z.ai · checked Nov 5, 2026
- zai-org/GLM-4.5-Air model card (opens in a new tab) huggingface.co · checked Nov 3, 2026
- zai-org/GLM-4.5-Air config.json (opens in a new tab) huggingface.co · checked Nov 5, 2026
- zai-org/GLM-4.5-Air files (opens in a new tab) huggingface.co · checked Oct 30, 2026
- zai-org/GLM-4.5-Air-FP8 files (opens in a new tab) huggingface.co · checked Oct 31, 2026
Zhipu AI’s official policies: data use, privacy, terms →
Some values on this page are still waiting to be checked against official pages: a filled dot is verified, a hollow one is not yet (with or without a source on file). How we verify data
Zhipu AI · China · z.ai
Related models
Snapshots, the same family and other models from this provider
- GLM-4-Air-250414
- GLM-5.3-FlashX-Legal
- GLM-5.3-Flash
- GLM-5.3
- GLM-5.2
- GLM-5.1
- GLM-5V-Turbo
- GLM-5-Turbo
- GLM-4.5-Air vs Muse Spark 1.2
- Claude Sonnet 5.5 vs GLM-4.5-Air
- Gemini 3.8 Flash vs GLM-4.5-Air
- GLM-4.5-Air vs GPT-5.6 Terra
- GLM-4.5-Air vs GPT-6.1 Sol
- GLM-4.5-Air vs Mistral Medium 3.5
- GLM-4.5-Air vs Muse Spark 1.3
- GLM-4.5-Air vs Qwen3.7-Plus