Hunyuan-A13B

Overview

Hunyuan-A13B is Tencent‘s open-source 80B-total / 13B-active Mixture-of-Experts model with a dual-mode Chain-of-Thought that adapts reasoning depth to task complexity. The technical report surfaces on HuggingFace papers as another credible open-weight MoE in the sub-15B-active tier, extending the fast/slow-thinking pattern already visible in recent closed models. Sits inside Tencent’s broader Hunyuan open-weights arc alongside Hy3 (295B / 21B-active, July 2026) and Tencent Hy4 (770B / 49B-active, August 2026).

Timeline

• 2026-09-24-AI-Digest — Hunyuan-A13B Technical Report surfaces on HuggingFace papers (arXiv:2609.27284, ▲4 upvotes) — Tencent‘s open-source 80B-total / 13B-active MoE with a dual-mode Chain-of-Thought adapting reasoning depth to task complexity, described as competitive across math, science, code, and general reasoning. Load-bearing framing: another credible open-weight MoE in the sub-15B-active tier, extending the fast/slow-thinking pattern already visible in recent closed models — the paper is a technical report surfacing rather than a fresh weight-drop event, and the model sits below Hy3 and Tencent Hy4 on the total-parameter scale.

Key Developments

1. Dual-Mode Chain-of-Thought With Adaptive Reasoning Depth (September 24, 2026): Technical report describes a runtime dual-mode CoT that adapts reasoning depth to task complexity — an open-weight analogue of the fast/slow-thinking pattern that has surfaced in recent closed models. Sub-15B-active tier makes the model directly relevant to on-desk / small-cluster inference budgets.

2. 80B-Total / 13B-Active MoE Under Tencent’s Hunyuan Line (September 24, 2026): Extends Tencent’s Hunyuan open-weights arc alongside Hy3 (295B / 21B-active, July 2026) and Tencent Hy4 (770B / 49B-active, August 2026), staking a sub-15B-active credible-frontier tier that the Chinese open-weights corpus has been widening throughout 2026.

Related

See also: Tencent, Hy3, Tencent Hy4, MOC - Open Source Models, MOC - Major Companies.