Grug 35B A3B — Hardware Requirements & GPU Compatibility
ChatReasoning
Grug 35B A3B is a 34.7B-parameter open language model from kai-os. It supports a context window of up to 262,144 tokens. At BF16 it needs about 69.71 GB of VRAM — see which GPUs and Macs can run it below.
Grug 35B A3B runs on a 96 GB+ Mac or a workstation GPU
Best local pick → Mac Studio (M3 Ultra, 96GB) · BF16 · 69.8 GB · ~50 tok/s
Macs from 96 GB (BF16 · ~50 tok/s)
Apache 2.0
Not set up for it?
Reference GPU · RTX 4090 24 GB
Estimates, ±20% · how we estimate
Run Grug 35B A3B
One recommended way. Every other method, OS and quant is in the step-by-step guide.
Recommended
vllm serve kai-os/Grug-35B-A3B --max-model-len 16384 --tensor-parallel-size 8vLLM repo with its own quantisation
Download: Weights on Hugging Face
Other ways
each opens step-by-stepRun in cloud
Fits on RTX PRO 6000 (96 GB) (26 GB headroom) · BF16
- Generation speed
- ~123 tok/s
- generation speed
- Cost per 1M output tokens
- $3.04
- per 1M output tokens
How Much VRAM Does Grug 35B A3B Need?
Select a quantization level; the listed hardware adheres to it.
| Quantization | VRAM | +32K | Runs on | |
|---|---|---|---|---|
| 69.7 GB | 70 GB | 16-bit · bfloat — Brain floating point 16 — preferred for training | Runs on RTX PRO 5000 Blackwell 72 GB |
Selected BF16
Brain floating point 1669.8GBat 16K context
- Weights
- 69.3 GB
- KV cache
- 0.2 GB
- Overhead
- 0.3 GB
Server-class: ~69.8 GB at 16K context · one RTX PRO 5000 Blackwell 72 GB
Which Hardware Can Run Grug 35B A3B?
BF16 · 69.7 GBGrug 35B A3B (BF16) requires 69.7 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 91+ GB is recommended. Using the full 262K context window can add up to 2.6 GB, bringing total usage to 72.3 GB. No consumer GPU has enough memory. 30 devices with unified memory can run Grug 35B A3B, including Mac Studio (M3 Ultra, 256GB), Mac Studio (M3 Ultra, 512GB), Mac Studio (M3 Ultra, 96GB).
96 GB
fits · 26.2 GB free · up to 256K ctx
3 GPUs · 2 Mac classes
RTX PRO 6000 Blackwell Max-Q Workstation Edition~123Simulate output on RTX PRO 6000 Blackwell Max-Q Workstation EditionRTX PRO 6000 Blackwell Workstation Edition~123Simulate output on RTX PRO 6000 Blackwell Workstation EditionRTX PRO 6000 Blackwell Server Edition~112Simulate output on RTX PRO 6000 Blackwell Server EditionM3 Ultra*96 GB~50Simulate output on Mac Studio (M3 Ultra, 96GB)M3 Max*96 GB~24Simulate output on MacBook Pro 14" M3 Max (96 GB)128–144 GB
fits · 58.2 GB free · up to 256K ctx
4 GPUs · 2 Mac classes · 4 device families
GH200 Grace Hopper Superchip~228Simulate output on GH200 Grace Hopper SuperchipH200 NVL~226Simulate output on H200 NVLH200 SXM~226Simulate output on H200 SXMM2 Ultra*128 GB~49Simulate output on Mac Pro M2 Ultra (128 GB)M5 Max128 GB~41Simulate output on Mac Studio M5 Max (128 GB)GB10~24Simulate output on NVIDIA DGX SparkJetson T5000 (Thor)~24Simulate output on NVIDIA Jetson AGX Thor Developer Kit
By VRAM tier · typical GPUs and Macs at BF16
Consumer GPUs (up to 32 GB) and Macs up to 64 GB: too small for this model (needs ≥ 69.8 GB)8 GB tier →12 GB tier →16 GB tier →24 GB tier →Macs →
Similar models
6| Model | Score | VRAM | tok/s | Why it is similar |
|---|---|---|---|---|
| Qwen3.6 35B A3BAlibaba | 106.8 | 22 GB | ~151 | same size |
| MiniCPM V 4.7 35B A3BOpenBMB | — | 22.1 GB | ~169 | same size |
| Ornith 1.5 35B A3Bornith-ai | — | 22 GB | ~151 | same size |
| Agents A1InternScience | — | 21.5 GB | ~171 | same size |
| Intern S2 PreviewInternLM | — | 73.2 GB | —needs > 24 GB | same size |
| EXAONE 4.5 33BLGAI-EXAONE | — | 23.9 GB | ~33 | same size |
Specifications
- Publisher
- kai-os
- Parameters
- 34.7B
- Active parameters
- 3.5B per token
- Architecture
- Qwen3_5MoeForCausalLM
- Context Length
- 262,144 tokens
- Vocabulary Size
- 248,320
- Release Date
- 2026-07-02
- License
- Apache 2.0
Frequently Asked Questions
How much VRAM does Grug 35B A3B need?
Grug 35B A3B requires 69.7 GB of VRAM at BF16. Full 262K context adds up to 2.6 GB (72.3 GB total).
VRAM = Weights + KV Cache + Overhead
Weights = 34.7B × 16 bits ÷ 8 = 69.3 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 ≈ 3 GB (at full 262K context)
VRAM usage by quantization
Can NVIDIA GeForce RTX 5090 run Grug 35B A3B?
No — Grug 35B A3B requires at least 69.7 GB at BF16, which exceeds the NVIDIA GeForce RTX 5090's 32 GB of VRAM.
Can I run Grug 35B A3B on a Mac?
Yes — Mac Studio (M3 Ultra, 96GB) and 20 other Macs can run Grug 35B A3B. Apple Silicon uses unified memory, so the model shares RAM with the system. At BF16 you need at least 69.7 GB of usable unified memory (RAM minus macOS overhead).
Can I run Grug 35B A3B locally?
Yes — Grug 35B A3B can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.
How fast is Grug 35B A3B?
At BF16, Grug 35B A3B can reach ~106 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 → 6.9 GB active ÷ (8000 × 0.65) = 1.336 ms, plus 40 layers × 0.055 ms = 2.200 ms, so 1000 ÷ 3.536 ms = ~283 tok/s
Estimated speed at BF16 (69.7 GB)
Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.
What's the download size of Grug 35B A3B?
At BF16, the download is about 69.32 GB.
Which GPUs can run Grug 35B A3B?
No single consumer GPU has enough VRAM to run Grug 35B A3B at BF16 (69.7 GB). Multi-GPU or professional hardware is required.
Which devices can run Grug 35B A3B?
30 devices with unified memory can run Grug 35B A3B at BF16 (69.7 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.