GLM 4.5 Air REAP 82B A12B — Hardware Requirements & GPU Compatibility
Chat
GLM 4.5 Air REAP 82B A12B is a 81.9B-parameter open language model from Cerebras in the GLM 4 family. It supports a context window of up to 131,072 tokens. At Q4_K_M it needs about 49.59 GB of VRAM — see which GPUs and Macs can run it below.
Usable on a 24 GB GPU
Runs on 24 GB (IQ2_XXS · ~131 tok/s · 16K)Macs from 32 GB (IQ2_S · ~44 tok/s)
MIT
Reference GPU · RTX 4090 24 GB
Estimates, ±20% · how we estimate
Run GLM 4.5 Air REAP 82B A12B
One recommended way. Every other method, OS and quant is in the step-by-step guide.
Recommended
ollama run hf.co/bartowski/cerebras_GLM-4.5-Air-REAP-82B-A12B-GGUF:Q4_K_MQ4_K_M needs offload on the RTX 4090 · 26.5 GB short · 16K context
Download: cerebras_GLM-4.5-Air-REAP-82B-A12B-Q4_K_M/cerebras_GLM-4.5-Air-REAP-82B-A12B-Q4_K_M-00001-of-00002.gguf56.57 GB + 1 partbartowski
Other ways
each opens step-by-stepRun in cloud
Fits on RTX PRO 6000 (96 GB) (45 GB headroom) · Q4_K_M
- Generation speed
- ~115 tok/s
- generation speed
- Cost per 1M output tokens
- $3.07
- per 1M output tokens
How Much VRAM Does GLM 4.5 Air REAP 82B A12B Need?
Select a quantization level; the listed hardware adheres to it.
| Quantization | VRAM | +32K | On RTX 4090 (reference) | |
|---|---|---|---|---|
| 35.3 GB | 37.2 GB | 2-bit · K-quant — 2-bit quantization with K-quant improvements | Does not fit: needs 36.2 GB | |
| 36.3 GB | 38.2 GB | 3-bit small — 3-bit small quantization | Does not fit: needs 37.2 GB | |
| 40.4 GB | 42.3 GB | 3-bit medium — 3-bit medium quantization | Does not fit: needs 41.3 GB | |
| 41.4 GB | 43.3 GB | 4-bit · legacy — 4-bit legacy quantization | Does not fit: needs 42.3 GB | |
| 49.6 GB | 51.5 GB | 4-bit medium — 4-bit medium quantization — most popular sweet spot | Does not fit: needs 50.5 GB | |
| 58.8 GB | 60.7 GB | 5-bit medium — 5-bit medium quantization — good quality/size tradeoff | Does not fit: needs 59.7 GB | |
| 68.0 GB | 70 GB | 6-bit, very good — 6-bit quantization, very good quality | Does not fit: needs 68.9 GB | |
| 82.4 GB | 84.3 GB | 8-bit, near-lossless — 8-bit quantization, near-lossless | Does not fit: needs 83.3 GB |
Selected Q4_K_M
4-bit medium quantization50.5GBat 16K context
- Weights
- 49.2 GB
- KV cache
- 1.0 GB
- Overhead
- 0.3 GB
Doesn't fit RTX 4090 (reference) · needs 26.5 GB more, IQ2_XXS fits
Which Hardware Can Run GLM 4.5 Air REAP 82B A12B?
Q4_K_M · 49.6 GBGLM 4.5 Air REAP 82B A12B (Q4_K_M) requires 49.6 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 65+ GB is recommended. Using the full 131K context window can add up to 8.1 GB, bringing total usage to 57.7 GB. No consumer GPU has enough memory. 52 devices with unified memory can run GLM 4.5 Air REAP 82B A12B, including Mac Studio (M3 Ultra, 256GB), Mac Studio (M3 Ultra, 512GB), Mac Pro M2 Ultra (64 GB).
≤ 20 GB
≤ 20 GB tier →doesn't fit · smallest quant needs 23.9 GB
32–40 GB
Q4_K_M · best fit IQ2_M (29.0 GB)
1 GPU · 3 Mac classes · 1 device family
RTX 5090~166Simulate output on RTX 5090M5 Max36 GB~46Simulate output on Mac Studio M5 Max (36 GB)M532 GB~21Simulate output on MacBook Air 13" M5 (32 GB)Orin~30Simulate output on NVIDIA Jetson AGX Orin 32GB64–80 GB
fits · 13.5 GB free · up to 128K ctx
3 Mac classes · 1 device family
M2 Ultra*64 GB~45Simulate output on Mac Pro M2 Ultra (64 GB)M5 Max64 GB~39Simulate output on Mac Studio M5 Max (64 GB)Orin~18Simulate output on NVIDIA Jetson AGX Orin 64GB
By VRAM tier · typical GPUs and Macs at Q4_K_M
Alternatives to GLM 4.5 Air REAP 82B A12B
GLM 4 family · 18 models →Similar models
5| Model | Score | VRAM | tok/s | Why it is similar |
|---|---|---|---|---|
| Kolibri 1 BF16Aleph-Alpha | — | 47.9 GB | —needs > 24 GB | same size |
| AliceAI Foundation 80B A3B Baseyandex | — | 49.9 GB | —needs > 24 GB | same size |
| Qwen3 Next 80B A3B InstructAlibaba | — | 22.9 GB | ~200 | same size |
| Hunyuan A13B InstructTencent | — | 50.7 GB | —needs > 24 GB | same size |
| Kolibri 1Aleph-Alpha | — | 47.9 GB | —needs > 24 GB | same size |
Specifications
- Publisher
- Cerebras
- Family
- GLM 4
- Parameters
- 81.9B
- Active parameters
- 11.9B per token
- Architecture
- Glm4MoeForCausalLM
- Context Length
- 131,072 tokens
- Vocabulary Size
- 151,552
- Release Date
- 2025-10-20
- License
- MIT
Frequently Asked Questions
How much VRAM does GLM 4.5 Air REAP 82B A12B need?
GLM 4.5 Air REAP 82B A12B requires 49.6 GB of VRAM at Q4_K_M, or 164.3 GB at BF16. Full 131K context adds up to 8.1 GB (57.7 GB total).
VRAM = Weights + KV Cache + Overhead
Weights = 81.9B × 4.8 bits ÷ 8 = 49.2 GB
KV Cache + Overhead ≈ 0.4 GB (at 2K context + ~0.3 GB framework)
Fit ratings and hardware model lists check this model with room for a 16K-token context, which needs a little more memory.
KV Cache + Overhead ≈ 8.5 GB (at full 131K context)
VRAM usage by quantization
Can NVIDIA GeForce RTX 4090 run GLM 4.5 Air REAP 82B A12B?
Yes, at IQ2_XXS (23.0 GB) or lower. Higher quantizations like IQ2_XS (25.0 GB) exceed the NVIDIA GeForce RTX 4090's 24 GB.
What's the best quantization for GLM 4.5 Air REAP 82B A12B?
For GLM 4.5 Air REAP 82B A12B, Q4_K_M (49.6 GB) offers the best balance of quality and VRAM usage. Q4_K_L (50.6 GB) provides better quality if you have the VRAM. The smallest option is IQ2_XXS at 23.0 GB.
VRAM requirement by quantization
★ Recommended — best balance of quality and VRAM usage.
Can I run GLM 4.5 Air REAP 82B A12B on a Mac?
Yes — Mac Studio M1 Max (32 GB) and 84 other Macs can run GLM 4.5 Air REAP 82B A12B, the smallest at IQ2_S. Apple Silicon uses unified memory, so the model shares RAM with the system. At Q4_K_M you need at least 49.6 GB of usable unified memory (RAM minus macOS overhead), from Mac Pro M2 Ultra (64 GB).
Can I run GLM 4.5 Air REAP 82B A12B locally?
Yes — GLM 4.5 Air REAP 82B A12B can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.
How fast is GLM 4.5 Air REAP 82B A12B?
At Q4_K_M, GLM 4.5 Air REAP 82B A12B can reach ~94 tok/s on AMD Instinct MI350X. Speed depends mainly on GPU memory bandwidth. Real-world results typically within ±20%.
tok/s = 1000 ÷ (active GB ÷ (bandwidth GB/s × efficiency) × 1000 + layers × routing ms)
Mixture-of-Experts: just the active experts get read for each token, along with a fixed routing cost per layer.
Example: NVIDIA B200 → 7.2 GB active ÷ (8000 × 0.65) = 1.38 ms, plus 46 layers × 0.055 ms = 2.53 ms, so 1000 ÷ 3.91 ms = ~256 tok/s
Estimated speed at Q4_K_M (49.6 GB)
Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.
What's the download size of GLM 4.5 Air REAP 82B A12B?
At Q4_K_M, the download is about 49.16 GB. The full-precision BF16 version is 163.86 GB. The smallest option (IQ2_XXS) is 22.53 GB.
Which GPUs can run GLM 4.5 Air REAP 82B A12B?
No single consumer GPU has enough VRAM to run GLM 4.5 Air REAP 82B A12B at Q4_K_M (49.6 GB). Multi-GPU or professional hardware is required.
Which devices can run GLM 4.5 Air REAP 82B A12B?
52 devices with unified memory can run GLM 4.5 Air REAP 82B A12B at Q4_K_M (49.6 GB), including ASUS Ascent GX10, Beelink GTR9 Pro (Ryzen AI Max+ 395, 128 GB), Framework Desktop (Ryzen AI Max+ 395, 128 GB), GMKtec EVO-X2 (Ryzen AI Max+ 395, 128 GB). Apple Silicon Macs use unified memory shared between CPU and GPU, making them well-suited for local LLM inference.