News
ReleasesOVHcloud unveils Olala mini, its first in-house LLMChinaA 9B model beats a 1T model on FinQA, trained with Crusoe Managed RLReleasesThe internal metaphysics of an LLMReleasesMapping Language Model Internal Representations to Physical Laws and Scientific ReasoningResearchCan China keep rogue AI in check? Meet the start-up testing the top modelsFrontierMicrosoft aims to challenge OpenAI, Anthropic with cheaper, specialized AI modelsOpen modelsMicrosoft rolls out new first-party models and updates to agentic AI continuous improvementFrontierMicrosoft unveils AI models for coding, cybersecurity as it takes on Anthropic, OpenAIFrontierGoogle is making it easier to opt out of Gemini AI model training – here’s howFrontierOpenAI Unveils New Astra AI Model After Scrapping Prior ReleaseReleasesOVHcloud unveils Olala mini, its first in-house LLMChinaA 9B model beats a 1T model on FinQA, trained with Crusoe Managed RLReleasesThe internal metaphysics of an LLMReleasesMapping Language Model Internal Representations to Physical Laws and Scientific ReasoningResearchCan China keep rogue AI in check? Meet the start-up testing the top modelsFrontierMicrosoft aims to challenge OpenAI, Anthropic with cheaper, specialized AI modelsOpen modelsMicrosoft rolls out new first-party models and updates to agentic AI continuous improvementFrontierMicrosoft unveils AI models for coding, cybersecurity as it takes on Anthropic, OpenAIFrontierGoogle is making it easier to opt out of Gemini AI model training – here’s howFrontierOpenAI Unveils New Astra AI Model After Scrapping Prior Release
EveryLLM
fnHunyuan-A13B

Hunyuan-A13B

Open-source MoE language model with 80 billion total parameters

Tencent released Hunyuan-A13B in June 2025 as an open-source large language model. It uses a mixture of experts design to balance high performance with lower computational costs. The system supports long contexts and complex reasoning tasks for developers.

Share on WhatsApp

  • Uses 13 billion active parameters out of 80 billion total
  • Supports a native context window of 256,000 tokens
  • Offers both fast and slow thinking reasoning modes
  • Released as open-source in June 2025 by Tencent
  • Optimized for agent tasks and tool calling

What it is

Hunyuan-A13B is a large language model developed by Tencent. It became available in June 2025 through open-source channels. The model contains 80 billion parameters in total but activates only 13 billion during operation. This structure aims to provide strong performance while reducing the computing power needed. It handles general tasks and advanced reasoning problems. Users can access it for research or application development where resources are limited.

How it came about

Tencent introduced this model to address the challenge of growing resource demands in artificial intelligence. The company released the pretrain and instruct versions on Hugging Face in late June 2025. They also provided technical reports and manuals to help users train and run the system. This release followed a trend of optimizing models for efficiency without sacrificing capability.

Inside the model

The architecture relies on a fine-grained mixture of experts approach. This means different parts of the model handle specific inputs, keeping active parameter counts low. It uses grouped query attention to speed up inference. The system supports a context window of 256,000 tokens, allowing it to process very long documents. Developers can choose between fast thinking for quick answers and slow thinking for detailed step-by-step reasoning. Quantization formats like FP8 and GPTQ-Int4 are supported for further efficiency.

What it can do

Benchmark results show competitive scores in mathematics, science, and coding. It performs well on agent-related tests such as BFCL-v3 and C3-Bench. In math evaluations like AIME 2024, it scored 87.3. The model also handles instruction following and text creation tasks. Comparisons suggest it rivals larger models in several categories while using fewer active resources. It supports tool calling for autonomous agent workflows.

How to use it

The weights are available for download on Hugging Face. Tencent offers an API and demo interfaces for testing. Deployment guides cover frameworks like TensorRT-LLM, vLLM, and SGLang. Docker images are provided for easier setup. Users can run the model locally or via cloud services that support these inference engines. The licence allows for open use and modification.

Questions people ask about Hunyuan-A13B

How many parameters does Hunyuan-A13B have?

The model has a total of 80 billion parameters. However, it uses a mixture of experts design, so only 13 billion parameters are active for any given token. This helps reduce computational load during inference.

Can I use this model for long documents?

Yes, it natively supports a context window of 256,000 tokens. This allows it to maintain stable performance when processing very long texts or complex inputs that require extensive background information.

Does it support reasoning steps?

Yes, it features hybrid reasoning support. You can enable slow thinking mode for detailed internal reasoning steps or switch to fast thinking mode to skip these steps for quicker responses. This is controlled via prompt tags or code settings.

Good reading and sources

    Where this page comes from

    There is no English Wikipedia article on Hunyuan-A13B yet, so the facts on this page come from the sources listed above: huggingface.co/tencent/Hunyuan-A13B-Instruct. Some are the subject's own pages, so its claims about itself are reported as its claims. EveryLLM wrote the text in its own words, with AI assistance, and compares it with the sources to make sure no passage is copied.

    It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-14.

    The weekly digest

    Every Monday: the model of the day, new pages on EveryLLM, and the conferences and anniversaries of the week ahead.