Atlas

Models

← All models

Hunyuan A13B Instruct

Tencent · 2025-06-27 · 80.4B parameters

Hunyuan-A13B-Instruct is Tencent Hunyuan's open-weight fine-grained MoE instruct model, with about 80B total parameters and 13B active parameters per token. It supports fast and slow thinking modes and a native 256K context window.

Benchmark scores

BenchmarkScore
SnakeBench Average Apples0.6
SnakeBench Best Apples2.0
SnakeBench Rating15.2
SnakeBench Total Apples14.0
SnakeBench Win Rate35.0
Loading Atlas data…