Whittle Qwen 3.8 35B A3B — Hardware Requirements & GPU Compatibility
ChatReasoningCodeMath
Whittle Qwen 3.8 35B A3B is a 35.6B-parameter open language model from logic65 in the Qwen 3.8 family. It supports a context window of up to 262,144 tokens. At Q4_K_M it needs about 21.72 GB of VRAM — see which GPUs and Macs can run it below.
Decent on a 24 GB GPU
Runs on 24 GB (Q4_K_M · ~200 tok/s · 32K)Macs from 24 GB (Q3_K_M · ~67 tok/s)
Apache 2.0
Reference GPU · RTX 4090 24 GB
Estimates, ±20% · how we estimate
Run Whittle Qwen 3.8 35B A3B
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
ollama run hf.co/logic65/Whittle-Qwen-3.8-35B-A3B-GGUF:Q4_K_MQ4_K_M fits fully on the RTX 4090 · 1.7 GB free · 16K context
Download: Whittle-Qwen-3.8-35B-A3B-Q4_K_M.gguf21.28 GBlogic65
Ways to runOther ways
each opens step-by-stepRun in cloud
Your own private copy · Q4_K_M · 32 GB GPU
- The easiest way to try this model
- Running in a few minutes, easy setup
- Chat right here or connect any app
$5+ bonus for new users
Runs on RTX PRO 4500 (32 GB) (9 GB headroom) · Q4_K_M
- Generation speed
- ~187 tok/s
- generation speed
- Cost per 1M output tokens
- $0.50
- per 1M output tokens
How Much VRAM Does Whittle Qwen 3.8 35B A3B Need?
Select a quantization level; the listed hardware adheres to it.
| Quantization | VRAM | +32K | On RTX 4090 (reference) | |
|---|---|---|---|---|
| 15.5 GB | 16.8 GB | 2-bit · K-quant — 2-bit quantization with K-quant improvements | Fits: 128K context · ~238 tok/s | |
| 17.7 GB | 19 GB | 3-bit medium — 3-bit medium quantization | Fits: 128K context · ~223 tok/s | |
| 21.7 GB | 23 GB | 4-bit medium — 4-bit medium quantization — most popular sweet spot | Fits: 32K context · ~200 tok/s | |
| 25.7 GB | 27 GB | 5-bit medium — 5-bit medium quantization — good quality/size tradeoff | Does not fit: needs 26.3 GB | |
| 29.7 GB | 31 GB | 6-bit, very good — 6-bit quantization, very good quality | Does not fit: needs 30.3 GB | |
| 36.0 GB | 37.2 GB | 8-bit, near-lossless — 8-bit quantization, near-lossless | Does not fit: needs 36.5 GB | |
| 71.5 GB | 72.8 GB | 16-bit · bfloat — Brain floating point 16 — preferred for training | Does not fit: needs 72.1 GB |
Selected Q4_K_M
4-bit medium quantization22.3GBat 16K context
- Weights
- 21.3 GB
- KV cache
- 0.7 GB
- Overhead
- 0.3 GB
Fits RTX 4090 (reference) · 1.7 GB free · ~200 tok/s
≥50 tok/s needs ≥158 GB/s · RTX 5090 ~265, RTX 3090 Ti ~200
Or .
Which Hardware Can Run Whittle Qwen 3.8 35B A3B?
Q4_K_M · 21.7 GBWhittle Qwen 3.8 35B A3B (Q4_K_M) requires 21.7 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 29+ GB is recommended. Using the full 262K context window can add up to 10.6 GB, bringing total usage to 32.4 GB. 7 GPUs can run it, including NVIDIA GeForce RTX 5090. 100 devices with unified memory can run Whittle Qwen 3.8 35B A3B, including Mac Studio (M3 Ultra, 256GB), Mac Studio (M3 Ultra, 512GB), Mac Studio M5 Max (36 GB).
20 GB
Q4_K_M · best fit Q3_K_M (18.3 GB)
1 GPU
RX 7900 XT~90Simulate output on RX 7900 XT32–40 GB
fits · 9.7 GB free · up to 128K ctx
1 GPU · 3 Mac classes · 4 device families
RTX 5090~265Simulate output on RTX 5090M5 Max36 GB~73Simulate output on Mac Studio M5 Max (36 GB)M532 GB~40Simulate output on MacBook Air 13" M5 (32 GB)Orin~63Simulate output on NVIDIA Jetson AGX Orin 32GBSnapdragon X Elite (X1E-80-100)~40Simulate output on Snapdragon X Elite Copilot+ PC
By VRAM tier · typical GPUs and Macs at Q4_K_M
Alternatives to Whittle Qwen 3.8 35B A3B
Qwen 3.8 family · 12 models →Similar models
6| Model | Score | VRAM | tok/s | Why it is similar |
|---|---|---|---|---|
| Nanbeige4.3 33BNanbeige | — | 23.7 GB | ~32 | same size |
| Sage A1.5InternScience | — | 22 GB | ~151 | same size |
| MiniCPM V 4.7 35B A3BOpenBMB | — | 22.1 GB | ~169 | same size |
| Llm Jp 4.1 33B Thinkingllm-jp | — | 22.5 GB | ~35 | same size |
| Agnes 3.0 FlashAgnes-AI | — | 22.1 GB | ~35 | same size |
| Ornith 1.5 35B A3Bornith-ai | — | 22 GB | ~151 | same size |
Specifications
- Publisher
- logic65
- Family
- Qwen 3.8
- Parameters
- 35.6B
- Active parameters
- 3B per token
- Architecture
- Qwen4ExpForCausalLM
- Context Length
- 262,144 tokens
- Vocabulary Size
- 248,320
- Release Date
- 2026-09-19
- License
- Apache 2.0
Frequently Asked Questions
How much VRAM does Whittle Qwen 3.8 35B A3B need?
Whittle Qwen 3.8 35B A3B requires 21.7 GB of VRAM at Q4_K_M, or 71.5 GB at BF16. Full 262K context adds up to 10.6 GB (32.4 GB total).
VRAM = Weights + KV Cache + Overhead
Weights = 35.6B × 4.8 bits ÷ 8 = 21.3 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 ≈ 11.1 GB (at full 262K context)
VRAM usage by quantization
Can NVIDIA GeForce RTX 4090 run Whittle Qwen 3.8 35B A3B?
Yes, at Q4_K_M (21.7 GB) or lower. Higher quantizations like Q5_K_M (25.7 GB) exceed the NVIDIA GeForce RTX 4090's 24 GB.
What's the best quantization for Whittle Qwen 3.8 35B A3B?
For Whittle Qwen 3.8 35B A3B, Q4_K_M (21.7 GB) offers the best balance of quality and VRAM usage. Q5_K_M (25.7 GB) provides better quality if you have the VRAM. The smallest option is Q2_K at 15.5 GB.
VRAM requirement by quantization
★ Recommended — best balance of quality and VRAM usage.
Can I run Whittle Qwen 3.8 35B A3B on a Mac?
Yes — Mac Mini M5 Pro (24 GB) and 106 other Macs can run Whittle Qwen 3.8 35B A3B, the smallest at Q3_K_M. Apple Silicon uses unified memory, so the model shares RAM with the system. At Q4_K_M you need at least 21.7 GB of usable unified memory (RAM minus macOS overhead), from Mac Studio M1 Max (32 GB).
Can I run Whittle Qwen 3.8 35B A3B locally?
Yes — Whittle Qwen 3.8 35B A3B can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.
How fast is Whittle Qwen 3.8 35B A3B?
At Q4_K_M, Whittle Qwen 3.8 35B A3B can reach ~119 tok/s on AMD Instinct MI350X. On NVIDIA GeForce RTX 4090: ~200 tok/s. 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 → 1.8 GB active ÷ (8000 × 0.65) = 0.35 ms, plus 40 layers × 0.055 ms = 2.20 ms, so 1000 ÷ 2.55 ms = ~392 tok/s
Estimated speed at Q4_K_M (21.7 GB)
Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.
What's the download size of Whittle Qwen 3.8 35B A3B?
At Q4_K_M, the download is about 21.34 GB. The full-precision BF16 version is 71.12 GB. The smallest option (Q2_K) is 15.11 GB.
Which GPUs can run Whittle Qwen 3.8 35B A3B?
7 consumer GPUs can run Whittle Qwen 3.8 35B A3B at Q4_K_M. Top options include AMD Radeon RX 7900 XTX, NVIDIA GeForce RTX 3090.
Which devices can run Whittle Qwen 3.8 35B A3B?
100 devices with unified memory can run Whittle Qwen 3.8 35B A3B at Q4_K_M (21.7 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.