GLM 4.1V 9B Thinking
10.3B parameters · 64k context · MIT · released 2025-06-28
visionreasoningchat
In short
GLM 4.1V 9B Thinking is an open model from Z.ai with 10.3B parameters. At mlx-8bit, GLM 4.1V 9B Thinking needs about 12 GB of memory at 8k context and 13 GB at 32k, counting the weights, the KV cache and runtime buffers. The cheapest hardware in Germany that runs GLM 4.1V 9B Thinking fully in memory at 32k context is the NVIDIA Tesla P40, 369 € at eBay, at about 16 tok/s (estimated).
Updated . Speeds are estimates; how they are worked out.
What to run it on in Germany
Fully in memory at 32k context, priced at retailers today.
Cheapest
NVIDIA Tesla P4016tok/sest.mlx-8bit
369 €
Best value
NVIDIA Tesla P4016tok/sest.mlx-8bit
369 €
Fastest
NVIDIA GeForce RTX 509085tok/sest.mlx-8bit
6.028 €
Mac
Mac Studio M4 Max 64GB27tok/sest.mlx-8bit
4.183 €
All hardware that runs it
Everything we price, best value first. Two cards split the model by layers.
| Hardware | Quant | Fits | Decode | Price | tok/s per 1,000 |
|---|---|---|---|---|---|
| NVIDIA Tesla P40 | mlx-8bit | fits | 16 tok/s | 369 € | 44.4 |
| AMD Radeon RX 9060 XT 16GB | mlx-8bit | fits | 13 tok/s | 550 € | 24.4 |
| Intel Arc A770 16GB | mlx-8bit | fits | 21 tok/s | 646 € | 33.0 |
| NVIDIA GeForce RTX 4060 Ti 16GB | mlx-8bit | fits | 14 tok/s | 650 € | 20.9 |
| NVIDIA GeForce RTX 5060 Ti 16GB | mlx-8bit | fits | 21 tok/s | 739 € | 28.7 |
| AMD Radeon RX 9070 XT | mlx-8bit | fits | 27 tok/s | 778 € | 34.4 |
| AMD Radeon RX 7600 XT | mlx-8bit | fits | 12 tok/s | 813 € | 14.9 |
| AMD Radeon RX 7900 XT | mlx-8bit | fits | 34 tok/s | 855 € | 39.2 |
| AMD Radeon RX 7800 XT | mlx-8bit | fits | 26 tok/s | 1.002 € | 26.2 |
| Intel Arc Pro B60 | mlx-8bit | fits | 17 tok/s | 1.074 € | 16.2 |
| NVIDIA GeForce RTX 4080 Super | mlx-8bit | fits | 35 tok/s | 1.100 € | 31.6 |
| NVIDIA GeForce RTX 5070 Ti | mlx-8bit | fits | 42 tok/s | 1.149 € | 36.9 |
| NVIDIA GeForce RTX 4070 Ti Super | mlx-8bit | fits | 32 tok/s | 1.247 € | 25.5 |
| Mac mini M4 32GB | mlx-8bit | fits | 6 tok/s | 1.400 € | 4.2 |
| NVIDIA GeForce RTX 5080 | mlx-8bit | fits | 45 tok/s | 1.500 € | 30.3 |
| AMD Radeon RX 7900 XTX | mlx-8bit | fits | 40 tok/s | 1.555 € | 25.9 |
| NVIDIA GeForce RTX 3090 | mlx-8bit | fits | 44 tok/s | 1.850 € | 23.9 |
| AMD Radeon AI PRO R9700 | mlx-8bit | fits | 27 tok/s | 1.867 € | 14.4 |
| NVIDIA GeForce RTX 3090 Ti | mlx-8bit | fits | 48 tok/s | 2.500 € | 19.0 |
| Mac mini M4 Pro 48GB | mlx-8bit | fits | 14 tok/s | 2.699 € | 5.0 |
| Mac mini M4 Pro 64GB | mlx-8bit | fits | 14 tok/s | 2.799 € | 4.8 |
| GMKtec EVO-X2 Ryzen AI Max+ 395 128GB | mlx-8bit | fits | 11 tok/s | 3.700 € | 2.9 |
| NVIDIA GeForce RTX 4090 | mlx-8bit | fits | 48 tok/s | 3.955 € | 12.0 |
| Mac Studio M4 Max 64GB | mlx-8bit | fits | 27 tok/s | 4.183 € | 6.5 |
| Mac Studio M4 Max 128GB | mlx-8bit | fits | 27 tok/s | 5.800 € | 4.7 |
| Mac Studio M3 Ultra 96GB | mlx-8bit | fits | 41 tok/s | 5.999 € | 6.8 |
| NVIDIA GeForce RTX 5090 | mlx-8bit | fits | 85 tok/s | 6.028 € | 14.1 |
| NVIDIA DGX Spark 128GB | mlx-8bit | fits | 13 tok/s | 8.313 € | 1.6 |
| ASUS Ascent GX10 128GB | mlx-8bit | fits | 13 tok/s | 8.623 € | 1.5 |
| Mac Studio M3 Ultra 256GB | mlx-8bit | fits | 41 tok/s | 15.555 € | 2.6 |
| Mac Studio M3 Ultra 512GB | mlx-8bit | fits | 41 tok/s | 18.995 € | 2.1 |
| NVIDIA RTX PRO 6000 Blackwell | mlx-8bit | fits | 85 tok/s | 19.076 € | 4.4 |
| Framework Desktop Ryzen AI Max+ 395 128GB | mlx-8bit | fits | 11 tok/s | – | – |
| Framework Desktop Ryzen AI Max+ 395 64GB | mlx-8bit | fits | 11 tok/s | – | – |
| NVIDIA GeForce RTX 3060 12GB | mlx-8bit | offload | 12 tok/s | 309 € | 38.2 |
| Intel Arc B580 | mlx-8bit | offload | 11 tok/s | 360 € | 31.7 |
| NVIDIA GeForce RTX 3080 10GB | mlx-8bit | offload | 11 tok/s | 500 € | 21.4 |
| NVIDIA GeForce RTX 5070 | mlx-8bit | offload | 16 tok/s | 699 € | 23.5 |
| NVIDIA GeForce RTX 4070 Super | mlx-8bit | offload | 14 tok/s | 707 € | 20.2 |
1 other card or machine we price cannot hold it at this context.
Memory needed
Weights, KV cache and runtime buffers at each context length (KV cache at fp16).
| Quant | File | @8k | @32k | @64k |
|---|---|---|---|---|
| mlx-8bitmlx | 11.0 GB | 12.0 GB | 13.1 GB | 14.5 GB |
Run it
Copy a command for the runtime you use.
- mlx-8bit
- LM Studio
lms get mlx-community/GLM-4.1V-9B-Thinking-8bit - MLX
mlx_lm.server --model mlx-community/GLM-4.1V-9B-Thinking-8bit
- LM Studio
Notes
Needs transformers 4.57.1 or later. It is a 9B bilingual (Chinese/English) vision-language reasoning model that handles images up to 4K resolution at any aspect ratio.
Questions
- How much VRAM does GLM 4.1V 9B Thinking need?
- At mlx-8bit, GLM 4.1V 9B Thinking needs about 12 GB of memory at 8k context and 13 GB at 32k, counting the weights, the KV cache and runtime buffers.
- What is the best GPU for running GLM 4.1V 9B Thinking locally?
- The cheapest hardware in Germany that runs GLM 4.1V 9B Thinking fully in memory at 32k context is the NVIDIA Tesla P40, 369 € at eBay, at about 16 tok/s (estimated). The fastest we price is the NVIDIA GeForce RTX 5090, at about 85 tok/s.
- Can I run GLM 4.1V 9B Thinking on a 16 GB GPU?
- Yes: at mlx-8bit it fits fully in 16 GB at 32k context. On the NVIDIA GeForce RTX 5080 it runs at about 45 tok/s (estimated).
- Can I run GLM 4.1V 9B Thinking on a 24 GB GPU?
- Yes: at mlx-8bit it fits fully in 24 GB at 32k context. On the NVIDIA GeForce RTX 3090 Ti it runs at about 48 tok/s (estimated).
- What licence is GLM 4.1V 9B Thinking released under?
- MIT, which allows commercial use.