Qwen3-Next-80B-A3B-Instruct GGUF size and VRAM requirements

License: apache-2.0 ⬇ 38,754 ❤ 199
Parameters79.67B
Context262,144

unsloth/Qwen3-Next-80B-A3B-Instruct-GGUF is a very large instruction-tuned chat model with 79.67 billion parameters, built on the qwen3next architecture. It is released under the apache-2.0 license and has been downloaded 38,754 times.

Qwen3-Next-80B-A3B-Instruct-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 48 transformer layers, a hidden size of 2,048, 16 attention heads. It uses a hybrid attention design: only 12 of its 48 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-Next-80B-A3B-Instruct-GGUF locally at a 4,096-token context, its quantized versions need between 19.95 GB (TQ1_0, lowest quality) and 149.4 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-Next-80B-A3B-Instruct-GGUF is about 0.09 GB, rising to roughly 6.0 GB at its full 262,144-token context. Shorter prompts free up memory for a higher-quality quantization.

For most users the best balance is IQ1_M, needing about 23.63 GB. That means unsloth/Qwen3-Next-80B-A3B-Instruct-GGUF fits entirely in the VRAM of a 24 GB GPU or larger, running fully on the GPU.

Available GGUF quantizations for unsloth/Qwen3-Next-80B-A3B-Instruct-GGUF include TQ1_0, IQ1_S, IQ1_M, IQ2_XXS, Q2_K, Q2_K_L, Q2_K_XL, IQ3_XXS, Q3_K_S, Q3_K_XL, Q3_K_M, IQ4_XS, IQ4_NL, Q4_0, Q4_K_S, Q4_K_XL, Q4_K_M, Q4_1, Q5_K_S, Q5_K_XL, Q5_K_M, Q6_K, Q6_K_XL, Q8_0, Q8_K_XL, 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.

→ Guide: How much VRAM do you need?

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 QualityWeights KVTotal Speed~Verdict
TQ1_0 2.05 Very low 19.06 GB 0.09 GB 19.95 GB 2.6 t/s Offload
IQ1_S 2.3 Very low 21.33 GB 0.09 GB 22.22 GB 2.3 t/s Offload
IQ1_M 2.45 Very low 22.73 GB 0.09 GB 23.63 GB 2.2 t/s Offload
IQ2_XXS 2.63 Low 24.41 GB 0.09 GB 25.3 GB Insufficient
Q2_K 2.93 Low 27.17 GB 0.09 GB 28.06 GB Insufficient
Q2_K_L 2.94 Low 27.24 GB 0.09 GB 28.13 GB Insufficient
Q2_K_XL 3.02 Low 28.06 GB 0.09 GB 28.95 GB Insufficient
IQ3_XXS 3.32 Fair 30.82 GB 0.09 GB 31.72 GB Insufficient
Q3_K_S 3.47 Fair 32.21 GB 0.09 GB 33.1 GB Insufficient
Q3_K_XL 3.58 Fair 33.19 GB 0.09 GB 34.08 GB Insufficient
Q3_K_M 3.85 Fair 35.67 GB 0.09 GB 36.56 GB Insufficient
IQ4_XS 4.28 Good 39.72 GB 0.09 GB 40.61 GB Insufficient
IQ4_NL 4.53 Good 42.01 GB 0.09 GB 42.9 GB Insufficient
Q4_0 4.55 Good 42.2 GB 0.09 GB 43.09 GB Insufficient
Q4_K_S 4.57 Good 42.38 GB 0.09 GB 43.28 GB Insufficient
Q4_K_XL 4.63 Good 42.9 GB 0.09 GB 43.8 GB Insufficient
Q4_K_M 4.87 Good 45.17 GB 0.09 GB 46.07 GB Insufficient
Q4_1 5.03 Very good 46.62 GB 0.09 GB 47.51 GB Insufficient
Q5_K_S 5.52 Very good 51.24 GB 0.09 GB 52.13 GB Insufficient
Q5_K_XL 5.69 Very good 52.77 GB 0.09 GB 53.67 GB Insufficient
Q5_K_M 5.7 Very good 52.91 GB 0.09 GB 53.8 GB Insufficient
Q6_K 6.58 Excellent 61.04 GB 0.09 GB 61.93 GB Insufficient
Q6_K_XL 6.88 Excellent 63.81 GB 0.09 GB 64.71 GB Insufficient
Q8_0 8.52 Excellent 78.99 GB 0.09 GB 79.88 GB Insufficient
Q8_K_XL 9.35 Excellent 86.68 GB 0.09 GB 87.57 GB Insufficient
BF16 16.01 Excellent 148.51 GB 0.09 GB 149.4 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-Next-80B-A3B-Instruct-GGUF?

unsloth/Qwen3-Next-80B-A3B-Instruct-GGUF is an instruction-tuned chat model with 79.67 billion parameters, based on the qwen3next architecture. It is released under the apache-2.0 license and distributed as GGUF files for local inference.

Is unsloth/Qwen3-Next-80B-A3B-Instruct-GGUF a Mixture-of-Experts (MoE) model?

Yes. unsloth/Qwen3-Next-80B-A3B-Instruct-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-Next-80B-A3B-Instruct-GGUF handle long context?

unsloth/Qwen3-Next-80B-A3B-Instruct-GGUF uses a hybrid attention design: only 12 of its 48 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.

How much VRAM do you need to run unsloth/Qwen3-Next-80B-A3B-Instruct-GGUF?

You need about 23.63 GB of VRAM to run unsloth/Qwen3-Next-80B-A3B-Instruct-GGUF entirely on the GPU using the IQ1_M quantization (at a 4,096-token context). Smaller quantizations lower the requirement at the cost of quality.

Can I run unsloth/Qwen3-Next-80B-A3B-Instruct-GGUF on an 8 GB GPU?

Partially. unsloth/Qwen3-Next-80B-A3B-Instruct-GGUF only fits on an 8 GB GPU by offloading part of it to system RAM (with IQ1_M), which runs but is slower.

Can I run unsloth/Qwen3-Next-80B-A3B-Instruct-GGUF on a 16 GB GPU?

Partially. unsloth/Qwen3-Next-80B-A3B-Instruct-GGUF only fits on a 16 GB GPU by offloading part of it to system RAM (with Q4_1), which runs but is slower.

Can I run unsloth/Qwen3-Next-80B-A3B-Instruct-GGUF on a 24 GB GPU?

Yes. With 24 GB of VRAM you can run unsloth/Qwen3-Next-80B-A3B-Instruct-GGUF fully on the GPU using IQ1_M (about 23.63 GB).

What context length does unsloth/Qwen3-Next-80B-A3B-Instruct-GGUF support?

unsloth/Qwen3-Next-80B-A3B-Instruct-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-Next-80B-A3B-Instruct-GGUF?

For unsloth/Qwen3-Next-80B-A3B-Instruct-GGUF, a strong default is Q4_K_M, which needs about 46.07 GB and keeps most of the quality while roughly halving the memory versus 8-bit. With VRAM to spare, Q5_K_M or Q6_K add a little more quality; if you are tight on memory, a smaller quantization still runs. Pick the highest quantization that fits your VRAM.