Tencent: Hunyuan A13B Instruct

Hunyuan-A13B-Instruct is a sparse Mixture-of-Experts chat model released by Tencent through its GitHub organization. Third-party listings describe it as carrying roughly eight billion total parameters while activating only about 1.3 billion at inference, a design that lets a modestly sized active pathway rival much larger dense models. This MoE shape is the foundation for its hybrid fast-thinking and slow-thinking modes, where lightweight responses and more deliberate reasoning can share the same underlying checkpoint.

Practically, the model is positioned for assistant-style workloads that benefit from long-context comprehension and tool use. A reseller listing highlights stable long-text understanding and claims agent capabilities verified by BFCL-v3 and τ-Bench, pointing at strong performance on function-calling and multi-step tasks. Through OpenAI-compatible gateways the model is exposed at a low per-token price tier, and the MoE design keeps active compute small, making it a reasonable fit for hosted agent pipelines, retrieval-heavy assistants, and latency-sensitive chat where deeper reasoning can be invoked on demand.