Qwen2-1.5B-Instruct

1.5B · Family: Qwen

Instruction-tuned 1.5B-parameter language model from Alibaba's Qwen2 series. Transformer with GQA and SwiGLU, 32K context window; weights under Apache 2.0.

✓ Active✓ Public access⚖ Open sourceLLMTool-using model📁 Qwen

Context window

32K

tokens

Parameters

1.5B

parameters

Release date

7 June 2024

🏢AlibabaProducer 🔬QwenResearch lab

Access:DownloadAPIDeployment:💻 Local☁ Cloud📱 On-device

Overview

Qwen2-1.5B-Instruct is an instruction-tuned (chat) language model with roughly 1.5B parameters, part of the Qwen2 series developed by Alibaba Cloud and released on 7 June 2024. It is a Transformer decoder using grouped-query attention (GQA), SwiGLU activation, rotary position embeddings (RoPE) and QKV bias. The model has 28 layers, a hidden size of 1536, 12 attention heads and 2 key-value heads (GQA).

It supports a context window of up to 32,768 tokens and was trained on multilingual data (English, Chinese and about 27 additional languages). The Instruct variant was produced through pretraining followed by supervised fine-tuning (SFT) and direct preference optimization (DPO). Weights are released under the Apache 2.0 license, allowing local deployment and further fine-tuning.

Its small size makes it suitable for local and edge deployment while still delivering solid results on language understanding, coding, mathematics and multilingual tasks compared with the previous generation Qwen1.5-1.8B-Chat.

Classification

LLMTool-using model

Family: Qwen

Access & deployment

DownloadAPI

LocalCloudOn-device

Weights: Open source

Key parameters

📏 Context: 32K

🧩 Parameters: 1.5B

✓ Tools · ✓ Fine-tuning

📥 Input: text

Technical specification

Context window

32K

tokens

Parameters

1.5B

parameters

License

Apache 2.0

Hardware requirements

About 1.5B parameters; in BF16 it needs roughly 3–4 GB of memory. Runs locally on a single consumer GPU or CPU; quantized (INT4/GGUF) it runs on edge devices.

Features:✓ Tool use✓ Fine-tuning

Modalities

⬇ Input

text

⬆ Output

textcode

Capabilities and applications

Native model capabilities

Coding

Generating, analysing and modifying code in many programming languages. Covers writing functions, debugging, refactoring, code review, and creating tests. Measured by benchmarks such as HumanEval and SWE-bench.

Category: coding

Reasoning

The model's ability to reason logically and solve complex problems.

Category: reasoning

Mathematical reasoning

The model's ability to solve mathematical tasks requiring multi-step reasoning — equations, proofs, combinatorics, geometry, calculus and competition-level problems.

Category: reasoning

Multilingual

Competence in many natural languages (from a few to over a hundred): understanding, generation, translation, and code-switching within a single conversation. Frontier models support a wide range of languages with comparable quality.

Category: language

Long context

Support for large context windows — tens to hundreds of thousands (or millions) of input tokens. Enables analysis of entire codebases, long documents, and many parallel conversations without losing earlier information. GPT-5.1 supports 400,000 tokens.

Category: language

Instruction following

Precisely following instructions contained in the prompt: response format, length, style, constraints (e.g. 'reply in six words'). GPT-5.1 significantly improved this capability compared to GPT-5.

Category: language

Function Calling

Category: planning

Benchmark results

5 benchmarks

MMLU

accuracy

52.4%