Qwen3.8-2.4T-A95B GGUF size and VRAM requirements
unsloth/Qwen3.8-2.4T-A95B-GGUF is a very large language model with 2446.18 billion parameters, built on the qwen35moe architecture. It is released under the other license and has been downloaded 16,556 times.
Qwen3.8-2.4T-A95B-GGUF is a Mixture-of-Experts model with 512 experts, of which 10 are active on each token. Routing only 10 of 512 experts makes it noticeably faster than a dense model of the same size — but every expert still has to be held in memory, so the numbers below are set by the full parameter count, not the active one. It is built from 92 transformer layers, a hidden size of 8,192, 64 attention heads. It uses a hybrid attention design: only 23 of its 92 layers keep a full KV cache, while the rest use linear attention with a constant-size state. Long contexts therefore cost far less memory than on a conventional model of this size.
To run unsloth/Qwen3.8-2.4T-A95B-GGUF locally at a 4,096-token context, its quantized versions need between 371.13 GB (Q1_0, lowest quality) and 4558.28 GB (BF16, highest quality) of memory, weights plus KV cache and a system margin included. Context length drives that memory directly: at 4,096 tokens the KV cache for Qwen3.8-2.4T-A95B-GGUF is about 0.36 GB, rising to roughly 23.0 GB at its full 262,144-token context. Shorter prompts free up memory for a higher-quality quantization.
Available GGUF quantizations for unsloth/Qwen3.8-2.4T-A95B-GGUF include Q1_0, IQ1_S, IQ1_M, IQ2_XXS, IQ2_XS, IQ3_XXS, IQ4_XS, Q8_0, BF16. The model supports a native context length of up to 262,144 tokens; a longer context grows the KV cache and the memory needed.
GGUF file size and memory by quantization
Compare real GGUF weight sizes, estimated KV cache and total memory for Q4, Q5, Q8 and every quantization published in this repository.
| Quant. | Bits | Quality | Weights | KV | Total | Speed~ | Verdict |
|---|---|---|---|---|---|---|---|
| Q1_0 | 1.3 | Very low | 369.97 GB | 0.36 GB | 371.13 GB | — | Insufficient |
| IQ1_S | 1.66 | Very low | 473.47 GB | 0.36 GB | 474.63 GB | — | Insufficient |
| IQ1_M | 1.84 | Very low | 525.22 GB | 0.36 GB | 526.38 GB | — | Insufficient |
| IQ2_XXS | 2.15 | Very low | 611.47 GB | 0.36 GB | 612.63 GB | — | Insufficient |
| IQ2_XS | 2.39 | Very low | 680.47 GB | 0.36 GB | 681.63 GB | — | Insufficient |
| IQ3_XXS | 3.12 | Low | 889.91 GB | 0.36 GB | 891.07 GB | — | Insufficient |
| IQ4_XS | 4.29 | Good | 1220.85 GB | 0.36 GB | 1222.01 GB | — | Insufficient |
| Q8_0 | 8.5 | Excellent | 2421.67 GB | 0.36 GB | 2422.83 GB | — | Insufficient |
| BF16 | 16.0 | Excellent | 4557.12 GB | 0.36 GB | 4558.28 GB | — | Insufficient |
KV cache computed from the model's exact architecture. Speed is a rough estimate bounded by memory bandwidth.
Frequently asked questions
What kind of model is unsloth/Qwen3.8-2.4T-A95B-GGUF?
unsloth/Qwen3.8-2.4T-A95B-GGUF is a language model with 2446.18 billion parameters, based on the qwen35moe architecture. It is released under the other license and distributed as GGUF files for local inference.
Is unsloth/Qwen3.8-2.4T-A95B-GGUF a Mixture-of-Experts (MoE) model?
Yes. unsloth/Qwen3.8-2.4T-A95B-GGUF is a Mixture-of-Experts model with 512 experts, of which 10 are activated per token. That makes it faster than a dense model of the same size, but all 512 experts must be loaded into memory, so the VRAM/RAM it needs is driven by the total parameter count, not the active one.
How does unsloth/Qwen3.8-2.4T-A95B-GGUF handle long context?
unsloth/Qwen3.8-2.4T-A95B-GGUF uses a hybrid attention design: only 23 of its 92 layers keep a full KV cache, and the rest use linear attention with a constant-size state. Long contexts cost far less memory than on a conventional model of this size.
Can I run unsloth/Qwen3.8-2.4T-A95B-GGUF on an 8 GB GPU?
No. unsloth/Qwen3.8-2.4T-A95B-GGUF does not fit on an 8 GB GPU, even with the smallest quantization and system RAM offloading.
Can I run unsloth/Qwen3.8-2.4T-A95B-GGUF on a 16 GB GPU?
No. unsloth/Qwen3.8-2.4T-A95B-GGUF does not fit on a 16 GB GPU, even with the smallest quantization and system RAM offloading.
Can I run unsloth/Qwen3.8-2.4T-A95B-GGUF on a 24 GB GPU?
No. unsloth/Qwen3.8-2.4T-A95B-GGUF does not fit on a 24 GB GPU, even with the smallest quantization and system RAM offloading.
What context length does unsloth/Qwen3.8-2.4T-A95B-GGUF support?
unsloth/Qwen3.8-2.4T-A95B-GGUF supports a native context length of up to 262,144 tokens. A longer context grows the KV cache, so it increases the memory needed to run the model.
What is the best quantization for unsloth/Qwen3.8-2.4T-A95B-GGUF?
For unsloth/Qwen3.8-2.4T-A95B-GGUF, higher bits-per-weight means better quality. A common sweet spot is a Q4_K_M or Q5_K_M quantization, which keeps most of the quality while roughly halving the memory versus 8-bit. Pick the highest quantization that still fits in your VRAM.