Activefrontierllm Open Source

Qwen2.5 110B

by Alibaba· Released September 2024· Cutoff June 2024

Qwen2.5 110B is Alibaba's largest open-source language model, featuring 110 billion parameters and a 128K token context window. It excels in coding, mathematics, and multilingual tasks, outperforming many proprietary models of similar size. The model is available under a permissive Apache 2.0 license.

Official Site API Docs 🤗 Hugging Face

Input cost

Free (open source)

Output cost

Free (open source)

Context window

128K tokens

Max output

8192 tokens

Modalities

text

Parameters

110B

License

Apache-2.0

Capabilities

Function CallingCode GenerationMultilingual SupportLong Context UnderstandingJSON ModeStreaming

Best For

Complex reasoning, coding, and multilingual tasks requiring a large context window.

Strengths

Strong performance on coding and math benchmarks (e.g., HumanEval 92.5%, GSM8K 96.4%)
128K token context window for long-document processing
Multilingual support covering over 29 languages
Open-source with Apache 2.0 license, enabling commercial use

Limitations

Large model size (110B) requires significant computational resources for inference
Not multimodal (text-only input/output)
May not match the latest proprietary models in some creative writing tasks
Knowledge cutoff is earlier than some newer models

Use Cases

Code generation and debugging

Mathematical problem solving

Long-document summarization and analysis

Multilingual translation and content creation

Chatbots and virtual assistants

Data extraction and structured output generation

Educational tutoring in STEM subjects

Improvements Over Previous Model

Parameter count increased from 72B to 110B, boosting overall performance
Context window expanded from 32K to 128K tokens
Improved coding benchmarks: HumanEval 92.5% vs 85% (Qwen2 72B)
Enhanced multilingual capabilities with support for 29+ languages
Better instruction following and reduced hallucination rates
Open-source license changed to Apache 2.0 from original Qwen license

Back to all models

Improvements Over Previous Model

Parameter count increased from 72B to 110B, boosting overall performance

Context window expanded from 32K to 128K tokens

Improved coding benchmarks: HumanEval 92.5% vs 85% (Qwen2 72B)

Enhanced multilingual capabilities with support for 29+ languages

Better instruction following and reduced hallucination rates

Open-source license changed to Apache 2.0 from original Qwen license