The Open-Source AI Stack
RSS
All models

Models · qwen

Qwen 2.5 72B Instruct

Open Alibaba · 2024-09-19 · Qwen License (Apache-2.0 for 72B variant requires agreement)

The 72B class set new open-weights leadership across most reasoning benchmarks at release, with an 18T-token pretrain. Released alongside coder, math, and 0.5B/1.5B/3B/7B/14B/32B sibling sizes the same day.

Cost

$0.36 / Mtok input
$0.40 / Mtok output

Together AI · as of 2026-05-21

via Artificial Analysis ↗

Speed

54.1 tok/sec output
1184 ms TTFT

Together AI · as of 2026-05-21

via Artificial Analysis ↗

Architecture

tokens in Embedding vocab 152,064 · qwen tokenizer × N layers Grouped-Query Attention RoPE context 131,072 tokens Dense MLP SwiGLU activation (standard) 72.7B active params Output projection tokens out
Schema-generated from data/models.yaml. Every label is auditable against the model's sources.

Specs

Architecture
dense
Total params
72.7B
Active params
72.7B
Context window
131K tokens
Attention
gqa
Position encoding
rope
Pretraining tokens
18.0T
Post-training
sft, dpo
OSI-approved
no
Data released
no
Training code
not released

Benchmarks

Each score carries the date it was published; we never infer or interpolate missing scores.

General reasoning

MMLU-Pro 72.0 as of 2026-05-21 source ↗
GPQA-Diamond 49.0 as of 2024-09-19 source ↗

Code

HumanEval 86.6 as of 2024-09-19 source ↗
LiveCodeBench 27.6 as of 2026-05-21 source ↗

Math

MATH 85.8 as of 2026-05-21 source ↗
AIME 2024 16.0 as of 2026-05-21 source ↗
AIME 2025 14.0 as of 2026-05-21 source ↗

Recommended use cases

  • multilingual deployment
  • long-context retrieval
  • fine-tuning base

Available quantizations

GGUF llama.cpp's container; the common local format, k-quants from Q2 to Q8. runs on llama.cpp, Ollama
AWQ Activation-aware 4-bit weight quantization for GPU serving. runs on vLLM, SGLang
GPTQ Post-training 4-bit weight quantization for GPU serving. runs on vLLM, SGLang, Transformers
EXL2 ExLlamaV2's variable-bitrate format for consumer GPUs. runs on ExLlamaV2
MLX Apple MLX 4/8-bit layout for Apple silicon. runs on Apple MLX
FP8 8-bit float, frequently a native release on Hopper / Blackwell GPUs. runs on vLLM, SGLang, TensorRT-LLM
bitsandbytes On-the-fly NF4 / INT8 weight quantization inside Transformers. runs on Transformers

Verified via the Hugging Face model tree ↗. Community quantizations change over time; the families shown are those with published weights at audit time.

Notable innovations

  • · 18T-token pretrain
  • · Full size ladder from 0.5B to 72B

Lineage

Final 72B-class dense Qwen before the Qwen 3 MoE pivot.

Derivatives

Sources