Qwen3.8 Flash Next Abliterated EXL3 2.50bpw — Hardware Requirements & GPU Compatibility
ChatVision
Qwen3.8 Flash Next Abliterated EXL3 2.50bpw is a 22.3B-parameter open language model from ghost-actual in the Qwen 3.8 family. It supports a context window of up to 262,144 tokens. At Q4_K_M it needs about 13.75 GB of VRAM — see which GPUs and Macs can run it below.
Runs well on a 24 GB GPU
Runs from 12 GB (Q3_K_M · 16K)Best on 24 GB (Q4_K_M · ~48 tok/s · 128K)Macs from 16 GB (Q3_K_M · ~13 tok/s)
Other
Reference GPU · RTX 4090 24 GB
Estimates, ±20% · how we estimate
Run Qwen3.8 Flash Next Abliterated EXL3 2.50bpw
One recommended way. Every other method, OS and quant is in the step-by-step guide.
Recommended
vllm serve ghost-actual/Qwen3.8-Flash-Next-Abliterated-EXL3-2.50bpw --max-model-len 16384vLLM repo with its own quantisation
Download: Weights on Hugging Face
Other ways
each opens step-by-stepRun in cloud
Fits on RTX A5000 (24 GB) (9 GB headroom) · Q4_K_M
- Generation speed
- ~36 tok/s
- generation speed
- Cost per 1M output tokens
- $1.22
- per 1M output tokens
How Much VRAM Does Qwen3.8 Flash Next Abliterated EXL3 2.50bpw Need?
Select a quantization level; the listed hardware adheres to it.
| Quantization | VRAM | +32K | On RTX 4090 (reference) | |
|---|---|---|---|---|
| 9.8 GB | 11.1 GB | 2-bit · K-quant — 2-bit quantization with K-quant improvements | Fits: 256K context · ~67 tok/s | |
| 11.2 GB | 12.5 GB | 3-bit medium — 3-bit medium quantization | Fits: 256K context · ~58 tok/s | |
| 13.8 GB | 15 GB | 4-bit medium — 4-bit medium quantization — most popular sweet spot | Fits: 128K context · ~48 tok/s | |
| 16.3 GB | 17.5 GB | 5-bit medium — 5-bit medium quantization — good quality/size tradeoff | Fits: 128K context · ~40 tok/s | |
| 18.8 GB | 20 GB | 6-bit, very good — 6-bit quantization, very good quality | Fits: 64K context · ~35 tok/s | |
| 22.6 GB | 23.9 GB | 8-bit, near-lossless — 8-bit quantization, near-lossless | Fits: 32K context · ~29 tok/s | |
| 44.9 GB | 46.2 GB | 16-bit · bfloat — Brain floating point 16 — preferred for training | Does not fit: needs 45.5 GB |
Selected Q4_K_M
4-bit medium quantization14.3GBat 16K context
- Weights
- 13.4 GB
- KV cache
- 0.7 GB
- Overhead
- 0.3 GB
Fits RTX 4090 (reference) · 9.7 GB free · ~48 tok/s
≥50 tok/s needs ≥1.06 TB/s · RTX 5090 ~85
Which Hardware Can Run Qwen3.8 Flash Next Abliterated EXL3 2.50bpw?
Q4_K_M · 13.8 GBQwen3.8 Flash Next Abliterated EXL3 2.50bpw (Q4_K_M) requires 13.8 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 18+ GB is recommended. Using the full 262K context window can add up to 10.6 GB, bringing total usage to 24.4 GB. 26 GPUs can run it, including NVIDIA GeForce RTX 5090, NVIDIA GeForce RTX 3090 Ti, NVIDIA GeForce RTX 5080. 124 devices with unified memory can run Qwen3.8 Flash Next Abliterated EXL3 2.50bpw, including Mac Studio (M3 Ultra, 256GB), Mac Studio (M3 Ultra, 512GB), Mac Mini M5 Pro (24 GB).
16–18 GB
16–18 GB tier →fits · 1.7 GB free · up to 32K ctx
18 GPUs · 1 Mac class
RTX 5080~45Simulate output on RTX 5080RTX 5070 Ti~42Simulate output on RTX 5070 TiRTX 4080 SUPER~35Simulate output on RTX 4080 SUPERM3 Pro*18 GB~8Simulate output on MacBook Pro 14" M3 Pro (18 GB)20 GB
fits · 5.7 GB free · up to 128K ctx
1 GPU
RX 7900 XT~35Simulate output on RX 7900 XT32–40 GB
fits · 17.7 GB free · up to 256K ctx
1 GPU · 3 Mac classes · 4 device families
RTX 5090~85Simulate output on RTX 5090M5 Max36 GB~23Simulate output on Mac Studio M5 Max (36 GB)M532 GB~8Simulate output on MacBook Air 13" M5 (32 GB)Orin~10Simulate output on NVIDIA Jetson AGX Orin 32GBSnapdragon X Elite (X1E-80-100)~6Simulate output on Snapdragon X Elite Copilot+ PC
By VRAM tier · typical GPUs and Macs at Q4_K_M
Alternatives to Qwen3.8 Flash Next Abliterated EXL3 2.50bpw
Qwen 3.8 family · 12 models →Similar models
6| Model | Score | VRAM | tok/s | Why it is similar |
|---|---|---|---|---|
| Magistral Small 2509Mistral AI | 85.6 | 18.1 GB | ~43 | same size |
| LFM2.5 24B A2BLiquid AI | — | 15 GB | ~221 | same size |
| InternVL3 5 GPT OSS 20B A4B Preview HFOpenGVLab | — | 13.6 GB | ~180 | same size |
| Tri 21B Thinktrillionlabs | — | 45.1 GB | —needs > 24 GB | same size |
| EuroLLM 22B Instruct 2512utter-project | — | 17.5 GB | ~46 | same size |
| ERNIE 4.5 21B A3B ThinkingBaidu | — | 14.3 GB | ~213 | same size |
Specifications
- Publisher
- ghost-actual
- Family
- Qwen 3.8
- Parameters
- 22.3B
- Architecture
- Qwen4ExpForConditionalGeneration
- Context Length
- 262,144 tokens
- Vocabulary Size
- 248,320
- Release Date
- 2026-09-18
- License
- Other
Frequently Asked Questions
How much VRAM does Qwen3.8 Flash Next Abliterated EXL3 2.50bpw need?
Qwen3.8 Flash Next Abliterated EXL3 2.50bpw requires 13.8 GB of VRAM at Q4_K_M, or 44.9 GB at BF16. Full 262K context adds up to 10.6 GB (24.4 GB total).
VRAM = Weights + KV Cache + Overhead
Weights = 22.3B × 4.8 bits ÷ 8 = 13.4 GB
KV Cache + Overhead ≈ 0.3 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 GB (at full 262K context)
VRAM usage by quantization
Can NVIDIA GeForce RTX 4090 run Qwen3.8 Flash Next Abliterated EXL3 2.50bpw?
Yes, at Q8_0 (22.6 GB) or lower. Higher quantizations like BF16 (44.9 GB) exceed the NVIDIA GeForce RTX 4090's 24 GB.
What's the best quantization for Qwen3.8 Flash Next Abliterated EXL3 2.50bpw?
For Qwen3.8 Flash Next Abliterated EXL3 2.50bpw, Q4_K_M (13.8 GB) offers the best balance of quality and VRAM usage. Q5_K_M (16.3 GB) provides better quality if you have the VRAM. The smallest option is Q2_K at 9.8 GB.
VRAM requirement by quantization
★ Recommended — best balance of quality and VRAM usage.
Can I run Qwen3.8 Flash Next Abliterated EXL3 2.50bpw on a Mac?
Yes — Mac Mini M2 Pro (16 GB) and 133 other Macs can run Qwen3.8 Flash Next Abliterated EXL3 2.50bpw, 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 13.8 GB of usable unified memory (RAM minus macOS overhead), from MacBook Pro 14" M3 Pro (18 GB).
Can I run Qwen3.8 Flash Next Abliterated EXL3 2.50bpw locally?
Yes — Qwen3.8 Flash Next Abliterated EXL3 2.50bpw can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.
How fast is Qwen3.8 Flash Next Abliterated EXL3 2.50bpw?
At Q4_K_M, Qwen3.8 Flash Next Abliterated EXL3 2.50bpw can reach ~349 tok/s on AMD Instinct MI350X. On NVIDIA GeForce RTX 4090: ~48 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 ÷ 13.75 × 0.65 = ~378 tok/s
Estimated speed at Q4_K_M (13.8 GB)
Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.
What's the download size of Qwen3.8 Flash Next Abliterated EXL3 2.50bpw?
At Q4_K_M, the download is about 13.36 GB. The full-precision BF16 version is 44.54 GB. The smallest option (Q2_K) is 9.46 GB.
Which GPUs can run Qwen3.8 Flash Next Abliterated EXL3 2.50bpw?
26 consumer GPUs can run Qwen3.8 Flash Next Abliterated EXL3 2.50bpw at Q4_K_M. Top options include AMD Radeon RX 7900 XTX, NVIDIA GeForce RTX 3090, NVIDIA GeForce RTX 3090 Ti, AMD Radeon RX 6800. 7 GPUs have plenty of headroom for comfortable inference.
Which devices can run Qwen3.8 Flash Next Abliterated EXL3 2.50bpw?
124 devices with unified memory can run Qwen3.8 Flash Next Abliterated EXL3 2.50bpw at Q4_K_M (13.8 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.