AWQ vs EXL2: which quantization format should you use?

Compared on hardware support, runtime, quality and what the 79-model index actually ships.

At a glance

AWQ vs EXL2: which quantization format should you use?
AWQEXL2
Runs onNVIDIA GPU (CUDA 11.8+)NVIDIA GPU (Ampere+ recommended)
RuntimevLLM · AutoAWQ · TGIExLlamaV2 · TabbyAPI
Best forHigh-throughput API serverMax single-GPU performance
Adoption estimate45%32%
Models in this index53 / 7921 / 79

AWQ

Activation-Aware Weight Quantization. High-accuracy INT4 for NVIDIA. Pairs perfectly with vLLM for server deployment.

Strengths

  • Best accuracy at 4-bit
  • Blazing fast with vLLM
  • Excellent batch throughput

Trade-offs

  • NVIDIA only
  • More setup than GGUF

EXL2

ExLlamaV2 format. Mixed-precision per-layer quantization — best accuracy-per-bit ratio. Fastest single-GPU inference available.

Strengths

  • Fastest GPU inference
  • Best accuracy per bit
  • Ultra-low 2bpw option

Trade-offs

  • NVIDIA only
  • Steeper learning curve

Models that ship both

These 16 models publish weights in both formats, so the two rows describe the same model rather than two different ones — the only place this comparison is measurable rather than editorial.

Models that ship both
ModelAWQ LevelQuality lossEXL2 LevelQuality loss
Qwen2.5 72B InstructAWQ INT43.5%EXL2 3.5bpw4.8%
Llama 3.1 70B InstructAWQ INT43.9%EXL2 3.5bpw5.2%
Qwen3 32B InstructAWQ INT43.6%EXL2 3.5bpw4.5%
Qwen2.5-Coder 32B InstructAWQ INT43.5%EXL2 3.5bpw4.5%
Mistral Small 24B InstructAWQ INT43.8%EXL2 4.65bpw2.2%
Devstral Small 1.1 24BAWQ INT44%EXL2 4.65bpw2.5%
Magistral Small 1.2 24BAWQ INT44%EXL2 4.65bpw2.5%
Qwen3 14B InstructAWQ INT43.5%EXL2 4.65bpw1.8%
Qwen2.5 14B InstructAWQ INT43.8%EXL2 4.65bpw2.1%
DeepSeek-R1-Distill-Qwen-14BAWQ INT43.6%EXL2 4.65bpw2%
Qwen3 8B InstructAWQ INT43.8%EXL2 4.65bpw2%
Llama 3.1 8B InstructAWQ INT44.5%EXL2 4.65bpw2.5%
DeepSeek-R1-Distill-Llama-8BAWQ INT43.5%EXL2 4.65bpw1.9%
Qwen2.5 7B InstructAWQ INT44.2%EXL2 4.65bpw2.2%
Qwen2.5-Coder 7B InstructAWQ INT44%EXL2 4.65bpw2%
Qwen3 4B InstructAWQ INT43.8%EXL2 4.65bpw2.1%

Quality loss is perplexity increase against the unquantized weights — lower is better, and only comparable within one model.

Which to choose