readygpu

GLM 4.5 Air

110.5B total, 17B active · 128k context · mit · released 2025-07-20

chatcodingagentic

What to run it on in the UK

Fully in memory at 8k context, priced at retailers today.

Something we price holds it fully in memory, but none reaches 15 tok/s, our bar for recommending it. The table below has the speeds.

All hardware that runs it

Everything we price, best value first. Two cards split the model by layers.

HardwareQuantFitsDecodePricetok/s per 1,000
Mac Studio M4 Max 128GBQ6_Kfits13 tok/s––
Mac mini M4 Pro 64GBQ2_Kfits13 tok/s––
Framework Desktop Ryzen AI Max+ 395 128GBQ6_Kfits5 tok/s––
Mac Studio M3 Ultra 512GBQ8_0fits16 tok/s––
NVIDIA DGX Spark 128GBQ8_0fits5 tok/s––
GMKtec EVO-X2 Ryzen AI Max+ 395 128GBQ6_Kfits5 tok/s––
Mac Studio M3 Ultra 256GBQ8_0fits16 tok/s––
Mac Studio M4 Max 64GBQ2_Kfits25 tok/s––
Mac Studio M3 Ultra 96GBQ4_K_Mfits25 tok/s––
NVIDIA RTX PRO 6000 BlackwellQ6_Kfits40 tok/s––
ASUS Ascent GX10 128GBQ8_0fits5 tok/s––
Framework Desktop Ryzen AI Max+ 395 64GBQ2_Kfits10 tok/s––
NVIDIA GeForce RTX 3090Q2_Koffload7 tok/s––
NVIDIA GeForce RTX 5090Q2_Koffload11 tok/s––
NVIDIA GeForce RTX 4090Q2_Koffload7 tok/s––
AMD Radeon RX 7900 XTXQ2_Koffload7 tok/s––
AMD Radeon AI PRO R9700Q2_Koffload9 tok/s––
AMD Radeon RX 7900 XTQ2_Koffload6 tok/s––
NVIDIA GeForce RTX 3090 TiQ2_Koffload7 tok/s––
Intel Arc Pro B60Q2_Koffload6 tok/s––
NVIDIA Tesla P40Q2_Koffload6 tok/s––

19 other cards and machines we price cannot hold it at this context.

Memory needed

Weights, KV cache and runtime buffers at each context length (KV cache at fp16).

QuantFile@8k@32k@64k@128k
Q4_K_Mgguf68.0 GB70.1 GB74.6 GB80.5 GB92.4 GB
Q2_Kgguf42.2 GB44.4 GB48.8 GB54.8 GB66.7 GB
Q3_K_Mgguf53.3 GB55.5 GB59.9 GB65.9 GB77.7 GB
mlx-4bitmlx56.0 GB58.2 GB62.6 GB68.6 GB80.5 GB
Q5_K_Mgguf77.8 GB80.0 GB84.4 GB90.4 GB102.2 GB
Q6_Kgguf92.2 GB94.4 GB98.8 GB104.8 GB116.7 GB
mlx-8bitmlx105.8 GB107.9 GB112.4 GB118.3 GB130.2 GB
Q8_0gguf109.4 GB111.6 GB116.0 GB122.0 GB133.8 GB

Run it

Copy a command for the runtime you use.

  • Q4_K_M
    • Ollamaollama run hf.co/unsloth/GLM-4.5-Air-GGUF:Q4_K_M
    • LM Studiolms get unsloth/GLM-4.5-Air-GGUF@q4_k_m
    • llama.cppllama-server -hf unsloth/GLM-4.5-Air-GGUF:Q4_K_M
  • Q2_K
    • Ollamaollama run hf.co/unsloth/GLM-4.5-Air-GGUF:Q2_K
    • LM Studiolms get unsloth/GLM-4.5-Air-GGUF@q2_k
    • llama.cppllama-server -hf unsloth/GLM-4.5-Air-GGUF:Q2_K
  • Q3_K_M
    • Ollamaollama run hf.co/unsloth/GLM-4.5-Air-GGUF:Q3_K_M
    • LM Studiolms get unsloth/GLM-4.5-Air-GGUF@q3_k_m
    • llama.cppllama-server -hf unsloth/GLM-4.5-Air-GGUF:Q3_K_M
  • mlx-4bit
    • LM Studiolms get mlx-community/GLM-4.5-Air-4bit
    • MLXmlx_lm.server --model mlx-community/GLM-4.5-Air-4bit
  • Q5_K_M
    • Ollamaollama run hf.co/unsloth/GLM-4.5-Air-GGUF:Q5_K_M
    • LM Studiolms get unsloth/GLM-4.5-Air-GGUF@q5_k_m
    • llama.cppllama-server -hf unsloth/GLM-4.5-Air-GGUF:Q5_K_M
  • Q6_K
    • Ollamaollama run hf.co/unsloth/GLM-4.5-Air-GGUF:Q6_K
    • LM Studiolms get unsloth/GLM-4.5-Air-GGUF@q6_k
    • llama.cppllama-server -hf unsloth/GLM-4.5-Air-GGUF:Q6_K
  • mlx-8bit
    • LM Studiolms get mlx-community/GLM-4.5-Air-8bit
    • MLXmlx_lm.server --model mlx-community/GLM-4.5-Air-8bit
  • Q8_0
    • Ollamaollama run hf.co/unsloth/GLM-4.5-Air-GGUF:Q8_0
    • LM Studiolms get unsloth/GLM-4.5-Air-GGUF@q8_0
    • llama.cppllama-server -hf unsloth/GLM-4.5-Air-GGUF:Q8_0