LANGUAGE MODEL Tencent Last updated:

Hunyuan-Large

Tencent's First Open Frontier MoE

Tencent's first major open-weight large model — a 389-billion- parameter mixture-of-experts that activates 52B per token. Released Nov 2024, it brought Tencent into the same league as DeepSeek and Alibaba on the open frontier.

Cost
Free
Open weights — self-host
Context
256K
Up to 256,000 tokens
How are Intelligence, Speed & Cost bucketed?
Intelligence and Speed buckets are percentile ranks on Artificial Analysis. Cost buckets are fixed dollar thresholds keyed off output-token price ($/M out).
Intelligence
  • Top 1%≤ 1%
  • Top 5%≤ 5%
  • Top 10%≤ 10%
  • Good≤ 25%
  • Medium≤ 50%
  • Below avg> 50%
Speed
  • Top 1%≥ 345 tok/s
  • Top 5%≥ 237 tok/s
  • Top 10%≥ 196 tok/s
  • Good≥ 146 tok/s
  • Medium≥ 90 tok/s
  • Slow< 90 tok/s
Cost
  • Freeopen weights · self-host
  • Low< $1 / M out
  • Moderate$1–5 / M out
  • High≥ $5 / M out

Why it matters

Made it three Chinese hyperscalers (Alibaba, DeepSeek/Tencent) shipping frontier-scale open MoE — a market structure unique to China through 2024–2026.

Core Capabilities

Long Documents
Handles entire codebases, books, and multi-doc RAG.
Generative
Produces images, video, audio, or other media.
Agent Workflows
Built for tool use and autonomous tasks.

Context Window

256k tokens
≈ 197+ pages
4k Chat 聊天
32k Long docs 长文档
128k Books 整本书
400k Multi-doc 多文档
1M Codebase 整个代码库
10M
256k

Availability

API
Not available
Product / App
Not available
Open Source
Released
Enterprise

Pricing Model

Free / self-host
Open weights — pay only for compute.
Self-host

Capability / Performance

Where this model sits relative to the middle 60% of models in the tree. All scores are 0–10 (higher is better).

Lower 20% Upper 80% This model
Context / memory
Context window size · log-scaled
6.0
9.0
7.0
Lower 20% 20th percentile — 20% of models score below this This model Where the current model lands Upper 80% 80th percentile — only 20% of models score above this Percentile boundaries are computed across every model in the tree that reports the underlying benchmark for each capability.

What it feels like

  • Largest open-source Transformer-based MoE at release: 389B total / 52B active per token
  • MMLU 88.4% — beat Llama 3.1 70B (79.3%) and matched Llama 3.1 405B at one-tenth the active params
  • GSM8K 92.8% — surpassed Llama 3.1 405B's 89% on grade-school math
  • Native 256K context window — long-doc QA at open-weights pricing
  • 1.5T of 7T training tokens were synthetic — first major open-weights model to lean this hard on synthetic
  • Mixed-expert routing innovations carried into later open MoE work (DeepSeek V3, etc.)

Best use cases

  • Self-hosted enterprise deployments where 389B MoE is acceptable
  • Long-context analysis (256K) on Chinese / English bilingual content
  • Distillation source — synthetic-data heritage makes outputs friendly for student training
  • Research on mixed-expert routing and synthetic-data scaling

Tools to try

Not ideal for

  • Edge / single-GPU deployments — multi-node serving required
  • Frontier reasoning by 2025 — DeepSeek V3, Qwen 3, Kimi K2 surpass it
  • Multimodal tasks (text-only at this generation; vision is in Hunyuan-V or Hunyuan-Video)