---
title: "Qwen2.5 Technical Report"
license: CC-BY-4.0
license_url: https://creativecommons.org/licenses/by/4.0/
license_note: 'paper is arXiv non-exclusive distribution license (verified on arXiv abs page 2026-11-12); this pointer note is our own words and is CC BY 4.0; we link to the paper rather than reproduce it'
source: https://iinventedlight.com/library/notes/2412-15115-qwen2.5-technical-report.md
source_license: "arXiv non-exclusive distribution license"
attribution: 'Pointer by iinventedlight (AI-assisted). Source: the Qwen team (authors listed on the arXiv abs page), Qwen2.5 Technical Report (arXiv non-exclusive distribution license). This note links to the paper rather than reproducing it.'
author: 'iinventedlight (AI-assisted)'
date: 2026-11-12
status: published
approved_by: "Cole Withington, 2026-11-12 14:44 PT, Librarian chat widget t51s316: \"Yes, publish all 100 second-list arXiv notes\""
checked_on: 2026-11-12
as_of: 2026-11-12
confidence: medium
untested: 'Pointer note only. Paper not reproduced. Abs page and license checked 2026-11-12.'
category: arxiv
topic_area: 'ai-research: foundations'
tags: arxiv, qwen, foundations, post-training, open-weight
arxiv_id: 2412.15115
alias: 2412-15115-qwen2.5-technical-report
type: pointer
format: pointer
sources:
  - id: S1
    title: "Qwen2.5 Technical Report"
    publisher: arXiv
    authors: "the Qwen team (authors listed on the arXiv abs page)"
    url: https://arxiv.org/abs/2412.15115
    version_read: "v2 (3 Jan 2025)"
    accessed: 2026-11-12
    license: "arXiv non-exclusive distribution license"
---
# Qwen2.5 Technical Report

## Summary
The [Qwen2.5 Technical Report](https://arxiv.org/abs/2412.15115) (2024) describes the Qwen2.5 model family, a widely used open-weight line just before Qwen3. Pretraining data is scaled from 7 trillion tokens in the prior generation to 18 trillion. Post-training includes supervised fine-tuning on more than a million samples and multi-stage reinforcement learning, with reported gains on long text, structured data, and instruction following.

What it introduced: a size range of open base and instruction models, including quantized builds, plus hosted mixture-of-experts variants called Qwen2.5-Turbo and Qwen2.5-Plus. The authors report that the open 72B instruction model is competitive with much larger models, including Llama 3 405B Instruct, and that the hosted variants compete on cost with GPT-4o-mini and GPT-4o. Qwen2.5 also became the base for specialized lines such as Qwen2.5-Math, Qwen2.5-Coder, QwQ, and multimodal models.

What it led to: a default open baseline through late 2024 and early 2025, and then Qwen3. Live notes: [2407-21783-llama-3-herd](https://iinventedlight.com/library/notes/2407-21783-llama-3-herd/). The Qwen3 report is a pointer note in this batch; Qwen2.5-Math is in batch H.

Where to read it: https://arxiv.org/abs/2412.15115 (v2, last revised 3 Jan 2025).

The paper is posted on arXiv under the arXiv non-exclusive distribution license. This note links to it rather than reproducing it. The note itself is CC BY 4.0.

License: this note is CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Pointer by iinventedlight (AI-assisted). Paper license: arXiv non-exclusive distribution; this note links to it and does not reproduce it.

---

Agents: [/agents.md](/agents.md) | [/.well-known/agent-library.json](/.well-known/agent-library.json).
