Cagliostro V3.5 — Hardware Requirements & GPU Compatibility
Chat
Cagliostro V3.5 is a 146M-parameter open language model from bench-labs. It supports a context window of up to 2,048 tokens. At BF16 it needs about 0.67 GB of VRAM — see which GPUs and Macs can run it below.
Runs great on a 24 GB GPU
Runs from 8 GB (BF16 · 2K)Best on 24 GB (BF16 · ~978 tok/s · 2K)Macs from 8 GB (BF16 · ~107 tok/s)
Apache 2.0
Reference GPU · RTX 4090 24 GB
Estimates, ±20% · how we estimate
Run Cagliostro V3.5
One recommended way. Every other method, OS and quant is in the step-by-step guide.Each way opens a step-by-step guide for your OS and quant.
Recommended
vllm serve bench-labs/cagliostro-v3.5 --max-model-len 2048vLLM repo with its own quantisation
Download: Weights on Hugging Face
Ways to runOther ways
each opens step-by-stepRun in cloud
Runs on RTX 3080 Ti (12 GB) (11 GB headroom) · BF16
- Generation speed
- ~885 tok/s
- generation speed
- Cost per 1M output tokens
- $0.038
- per 1M output tokens
How Much VRAM Does Cagliostro V3.5 Need?
Select a quantization level; the listed hardware adheres to it.
| Quantization | VRAM | +32K | On RTX 4090 (reference) | |
|---|---|---|---|---|
| 0.7 GB | — | 16-bit · bfloat — Brain floating point 16 — preferred for training | Fits: 2K context · ~978 tok/s |
Selected BF16
Brain floating point 160.7GBat 2K context
- Weights
- 0.3 GB
- KV cache
- 0.1 GB
- Overhead
- 0.3 GB
Fits RTX 4090 (reference) · 23.3 GB free · ~978 tok/s
≥50 tok/s needs ≥51 GB/s · RTX 5090 ~1739, RTX 3090 Ti ~978
Which Hardware Can Run Cagliostro V3.5?
BF16 · 0.7 GBCagliostro V3.5 (BF16) requires 0.7 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 1+ GB is recommended. 52 GPUs can run it, including NVIDIA GeForce RTX 5090, NVIDIA GeForce RTX 3090 Ti. 163 devices with unified memory can run Cagliostro V3.5, including Mac Studio (M3 Ultra, 256GB), Mac Studio (M3 Ultra, 512GB).
≤ 11 GB
≤ 11 GB tier →fits · 7.3 GB free · up to 2K ctx
16 GPUs · 1 Mac class · 2 device families
RTX 3080~738Simulate output on RTX 3080RTX 2080 Ti~598Simulate output on RTX 2080 TiRTX 3070 Ti~590Simulate output on RTX 3070 TiM3*8 GB~107Simulate output on iMac 24" M3 (8 GB)Orin Nano~99Simulate output on NVIDIA Jetson Orin Nano 8GB (Super)iPhone~71Simulate output on iPhone 1720 GB
fits · 19.3 GB free · up to 2K ctx
1 GPU
RX 7900 XT~716Simulate output on RX 7900 XT32–40 GB
fits · 31.3 GB free · up to 2K ctx
1 GPU · 3 Mac classes · 4 device families
RTX 5090~1739Simulate output on RTX 5090M5 Max36 GB~481Simulate output on Mac Studio M5 Max (36 GB)M532 GB~161Simulate output on MacBook Air 13" M5 (32 GB)Orin~199Simulate output on NVIDIA Jetson AGX Orin 32GBSnapdragon X Elite (X1E-80-100)~131Simulate output on Snapdragon X Elite Copilot+ PC
By VRAM tier · typical GPUs and Macs at BF16
Fine-tunes & variants(1)
Specifications
- Publisher
- bench-labs
- Parameters
- 146M
- Architecture
- CagliostroForCausalLM
- Context Length
- 2,048 tokens
- Vocabulary Size
- 32,768
- Release Date
- 2026-10-04
- License
- Apache 2.0
Frequently Asked Questions
How much VRAM does Cagliostro V3.5 need?
Cagliostro V3.5 requires 0.7 GB of VRAM at BF16.
VRAM = Weights + KV Cache + Overhead
Weights = 146M × 16 bits ÷ 8 = 0.3 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.
VRAM usage by quantization
Can I run Cagliostro V3.5 on a Mac?
Yes — MacBook Air 13" M3 (8 GB) and 145 other Macs can run Cagliostro V3.5. Apple Silicon uses unified memory, so the model shares RAM with the system. At BF16 you need at least 0.7 GB of usable unified memory (RAM minus macOS overhead).
Can I run Cagliostro V3.5 locally?
Yes — Cagliostro V3.5 can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.
How fast is Cagliostro V3.5?
At BF16, Cagliostro V3.5 can reach ~7164 tok/s on AMD Instinct MI350X. On NVIDIA GeForce RTX 4090: ~978 tok/s. Speed depends mainly on GPU memory bandwidth. Real-world results typically within ±20%.
tok/s = (bandwidth GB/s ÷ model GB) × efficiency
Example: NVIDIA B200 → 8000 ÷ 0.67 × 0.65 = ~7761 tok/s
Estimated speed at BF16 (0.7 GB)
Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.
What's the download size of Cagliostro V3.5?
At BF16, the download is about 0.29 GB.
Which GPUs can run Cagliostro V3.5?
52 consumer GPUs can run Cagliostro V3.5 at BF16. Top options include AMD Radeon RX 6700 XT, AMD Radeon RX 6800, AMD Radeon RX 6800 XT. 52 GPUs have plenty of headroom for comfortable inference.
Which devices can run Cagliostro V3.5?
163 devices with unified memory can run Cagliostro V3.5 at BF16 (0.7 GB), including AMD Ryzen AI 9 HX 370 (Strix Point) Laptop, ASUS Ascent GX10, Beelink GTR9 Pro (Ryzen AI Max+ 395, 128 GB), Framework Desktop (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.