Qwen2.5-1MOpen Weight

Released

Qwen2.5-1M is an AI model released by Qwen on Monday, Jan 27 2025, 60 days after QwQ-32B-Preview. It is an open-weight model — the trained weights are available to download and run. It was released in 7B and 14B parameter sizesThe learned values that shape how a model responds. B means billion and T means trillion. For mixture-of-experts models, the total count is larger than the number active for each token. Parameter count alone does not measure quality., with a 1M token context windowHow much information the model can work with in one request, measured in tokens — pieces of words or other data. This includes your question, attachments, conversation and reply. Limits can vary by model variant and provider.. Benchmark results (shown below) cover LiveCodeBench, HumanEval, IFEval, MATH, MMLU-Pro, and GSM8K.

Benchmarks

Coding

LiveCodeBenchCompetitive coding — Coding problems published so recently the AI can't have seen them in training — a contamination-free test of raw programming skill. Higher is better.
38.6%
#69 of 86
HumanEvalFunction synthesis — 164 small Python problems: the AI is given a function's description and has to write the working function. This was the coding benchmark of the GPT-3.5 and GPT-4 era, before the field moved to fixing real bugs in real repositories. Higher is better.
88.4%
#20 of 88

Agentic & tool use

IFEvalAgentic tasks — following verifiable instructions in generated responses
84.3%
#12 of 24

Reasoning & science

MATHReasoning — accuracy on the original competition mathematics test set
79.5%
#11 of 60
GSM8KGrade-school math — Grade-school maths word problems that take a few steps of arithmetic to work through. It separated the models of 2022 and 2023 sharply, then saturated. One caveat on the historical numbers: OpenAI included part of the GSM8K training set in GPT-4's pre-training mix, so GPT-4's score is not a clean few-shot result. Higher is better.
94.8%
#9 of 56

Knowledge work

MMLU-ProKnowledge — accuracy on the ten-option MMLU-Pro benchmark
63.3%
#54 of 58

Compare Qwen2.5-1M with

Qwen2.5-1M

About

Qwen2.5-1M shipped on January 27, 2025 with two instruction-tuned checkpoints, in 7B and 14B sizes. It brought the million-token context that Alibaba had introduced on the Turbo API to downloadable Qwen weights.

Qwen published a vLLM-based inference implementation using sparse attention to reduce the cost of processing long inputs. Its short-context evaluation recorded 88.4 on HumanEval for the 14B checkpoint, showing how the long-context tuning affected a familiar coding task.

Frequently asked questions

Qwen2.5-1M was released by Qwen on Monday, Jan 27 2025.

All Qwen releases

42 tracked