Activereasoningllm Open Source

Qwen3 8B

by Alibaba· Released May 2025· Cutoff April 2025

Qwen3 8B is a 8-billion-parameter large language model from Alibaba's Qwen3 series, featuring a Mixture-of-Experts (MoE) architecture with 2.2B activated parameters. It supports extended thinking (reasoning) and is optimized for efficiency, outperforming similarly sized dense models on benchmarks like MMLU and coding tasks.

Official Site API Docs 🤗 Hugging Face 📄 Research Paper

Input cost

Free (open source)

Output cost

Free (open source)

Context window

32K tokens

Max output

8192 tokens

Modalities

text

Parameters

8B (2.2B activated)

License

Apache-2.0

Capabilities

Extended Thinking (Reasoning)Function CallingCode GenerationMultilingual SupportStreamingJSON Mode

Best For

Efficient reasoning and coding tasks with a balance of performance and low computational cost.

Strengths

Strong reasoning capabilities for its size due to MoE architecture
High efficiency with only 2.2B activated parameters
Competitive performance on benchmarks like MMLU and HumanEval
Multilingual support including English, Chinese, and others

Limitations

Smaller context window (32K) compared to larger Qwen3 models
Not multimodal (text-only)
May underperform on complex long-context tasks
Limited to 8B total parameters, less capable than larger variants

Use Cases

Chatbots and virtual assistants

Code generation and debugging

Reasoning and problem-solving tasks

Educational tools and tutoring

Content summarization and translation

Data extraction and analysis

Automated customer support

Improvements Over Previous Model

Introduced MoE architecture with 2.2B activated parameters vs dense 7B in Qwen2.5-7B
Added extended thinking (reasoning) capability not present in Qwen2.5
Improved MMLU score from ~72% (Qwen2.5-7B) to ~75% (Qwen3-8B)
Better coding performance: HumanEval score increased from ~75% to ~80%
Supports longer output (8K tokens) vs 2K in Qwen2.5-7B
Multilingual support expanded beyond Chinese and English

Back to all models

Activereasoningllm Open Source

Qwen3 8B

by Alibaba· Released May 2025· Cutoff April 2025

Official Site API Docs 🤗 Hugging Face 📄 Research Paper

Input cost

Free (open source)

Output cost

Free (open source)

Context window

32K tokens

Max output

8192 tokens

Modalities

text

Parameters

8B (2.2B activated)

License

Apache-2.0

Capabilities

Extended Thinking (Reasoning)Function CallingCode GenerationMultilingual SupportStreamingJSON Mode

Best For

Efficient reasoning and coding tasks with a balance of performance and low computational cost.

Strengths

Strong reasoning capabilities for its size due to MoE architecture
High efficiency with only 2.2B activated parameters
Competitive performance on benchmarks like MMLU and HumanEval
Multilingual support including English, Chinese, and others

Limitations

Smaller context window (32K) compared to larger Qwen3 models
Not multimodal (text-only)
May underperform on complex long-context tasks
Limited to 8B total parameters, less capable than larger variants

Use Cases

Chatbots and virtual assistants

Code generation and debugging

Reasoning and problem-solving tasks

Educational tools and tutoring

Content summarization and translation

Data extraction and analysis

Automated customer support

Improvements Over Previous Model

Introduced MoE architecture with 2.2B activated parameters vs dense 7B in Qwen2.5-7B
Added extended thinking (reasoning) capability not present in Qwen2.5
Improved MMLU score from ~72% (Qwen2.5-7B) to ~75% (Qwen3-8B)
Better coding performance: HumanEval score increased from ~75% to ~80%
Supports longer output (8K tokens) vs 2K in Qwen2.5-7B
Multilingual support expanded beyond Chinese and English

Back to all models