Model
Model explorer

Hunyuan-A13B

OPEN
Tencent · Hunyuan family · released Jun 27, 2025

Open-weights fine-grained MoE (80B total/13B active) with dual fast/slow reasoning, GQA and 256K context; leads open agentic benchmarks like BFCL-v3 (78.3).

ReasoningCodingVisionFunction callingTool useAgentic
1517.1
Elo · rank #118
Parameters
80B
Active params
13B (MoE)
Context
256K tokens
Architecture
Fine-grained sparse MoE Transformer, 1 shared + 64 routed experts, 8 active/token (80B total / 13B active)
License
Tencent Hunyuan Community License Agreement
Languages
—
API price (in/out)
$0.07 / $0.3
Modalities
text
Benchmark results
Bar shows position within the tracked field; marker = field best
AIMEMath87.3%#55
best: GPT-5.2 · 100.0%
BIG-Bench HardReasoning89.1%#8
best: ERNIE 4.5 300B-A47B · 94.3%
BFCL v3Agents78.3%#1
best: this model · 78.3%
DROPReasoning91.1#6
best: Hunyuan-T1 · 93.1
GPQA DiamondReasoning71.2%#121
best: GPT-5.6 · 94.6%
IFEvalReasoning84.7%#66
best: Gemma 4 26B A4B · 98.5%
LiveCodeBenchCoding63.9%#89
best: DeepSeek-V4-Pro (Think Max) · 93.5%
MATH-500Math94.3%#36
best: GPT-5 · 99.4%
tau-benchAgents54.7%#18
best: Claude Opus 4.1 · 82.4%
Run it locally
VRAM @ Q4
44 GB
VRAM @ FP16
160 GB
Fits on (Q4)
M3 Max 128GBM3 Ultra 512GBA100 80GBH100 80GBH200 141GBB200 192GB
80B total params exceed a single RTX 4090's 24GB VRAM at 4-bit despite only 13B active; needs multi-GPU or CPU offload to serve.
Quantizations
GGUF Q4 · GPTQ-Int4 · FP8
Hunyuan family
Elo progression across releases
Hunyuan (1st-gen)Sep 20231000.0Hunyuan-Large (A52B)Nov 20241072.5Hunyuan-TurboSFeb 20251306.0Hunyuan-T1Mar 20251631.0Hunyuan-A13BJun 20251517.1Hy3Jul 20262485.4
Resources
HFtencent/Hunyuan-A13B-InstructGHTencent-Hunyuan/Hunyuan-A13BAPIAvailable via Tencent API
API price $0.07/$0.3 · each benchmark row carries its own source badge (see methodology)