Tencent · Hunyuan · Reviewed 2026-09-11

Tencent Hunyuan-Large

Tencent's flagship open-source Mixture-of-Experts model with 389B total parameters, 52B active parameters, and native 256K context support.

Overall · 0–100—

Status: active
Released: 2024-11-05
Context: 262,144 tokens
Parameters: 389B total · 52B active
Inputs: Text
Type: text
Max output: 8,192 tokens
API model ID: hunyuan-large
Reasoning: —
Knowledge cutoff: 2024-09
Training cutoff: —
Weights / license: Available · Tencent Hunyuan Community License
Price / 1M tokens: —
Cached input: —

Independent benchmark scores

No snapshot score

Coding

—

No comparable score for this task lens in the selected independent snapshot.

No snapshot score

Reasoning

—

No comparable score for this task lens in the selected independent snapshot.

No snapshot score

Research

—

No comparable score for this task lens in the selected independent snapshot.

No snapshot score

Multimodal

—

No comparable score for this task lens in the selected independent snapshot.

No snapshot score

Agentic

—

No comparable score for this task lens in the selected independent snapshot.

Raw benchmark results

Source-backed rows with the original evaluation context preserved.

0 results

No verified raw benchmark rows are registered for this model yet.

Strengths

Open-source MoE model with 52B activated parameters per token
256K long context with KV cache compression (CLA/GQA)
Strong academic and code performance outperforming Llama 3.1-70B

Limitations

Text-only base model (multimodal requires Hunyuan-Large-Vision)
Large hardware footprint for full FP16/BF16 deployment

Model evidence · paper · Checked 2026-09-11

Tencent Hunyuan-Large: An Open-Source MoE Model with 52B Activated Parameters

Official technical report for Tencent Hunyuan-Large covering 389B MoE architecture, 256K context, and evaluations.

Model evidence · official · Checked 2026-09-11

Tencent-Hunyuan-Large model card

Official open-weight model repository under Tencent Hunyuan Community License.

Capability Index source

Artificial Analysis Intelligence Index v4.3.2

Underlying raw metrics and exact model configurations remain authoritative.

Capability Index source

Artificial Analysis methodology

Underlying raw metrics and exact model configurations remain authoritative.

Task lenses preserve raw scores. Radar axes use score / current leader × 100; overall is the equal-weight average of all five indices. Missing evidence is not a zero or a low rank.