↓ Skip to main content
9 – 17 UHR +49 8031 3508270 LUITPOLDSTR. 9, 83022 ROSENHEIM
DE / EN
LLM Tencent China

Tencent Hunyuan (Hy3)

Tencent Hunyuan Hy3 and Hy4-Preview: open-source MoE models under Apache 2.0 without EU restrictions. Benchmarks, API pricing and GDPR-compliant self-hosting.

License Apache 2.0 (Hy3 full release since 6 July 2026, Hy4-Preview since 28 August 2026; the April 2026 Hy3 preview licence excluded the EU)
GDPR Hosting Available
Context 256000 Tokens
Modality Text, Code → Text, Code

Versions

Overview of available model variants

ModelReleaseEUStrengthsWeaknessesStatus
Hy4-Preview
August 28, 2026
MoE architecture: 770B total, 49B active parameters 1M token context per the model card Apache 2.0 licence even for the preview – unlike the Hy3 preview from April 2026 Scored 2.99 out of 4.00 in Tencent's internal blind evaluation (163 experts, 203 engineering tasks), narrowly ahead of GLM-5.3 (2.92) and Kimi K3 (2.94) Additional FP8-quantised variant (tencent/Hy4-preview-FP8) for more efficient self-hosting Already integrated into WorkBuddy, CodeBuddy, Yuanbao and ima on launch day
Still preview status – per Tencent not a final release, behaviour may still change The internal comparison evaluation comes from Tencent itself, no independent third-party confirmation known Known issues per the model card: reasoning longer than necessary on complex tasks and a tendency to over-verify its own work API via Tencent Cloud TokenHub (Chinese provider, no EU region) – for EU companies only self-hosting is advisable
Current
Hy3 Recommended
6 July 2026
MoE architecture: 295B total, only 21B active parameters (192 experts, top-8) 256K token context Hybrid fast/slow thinking (toggleable reasoning) Multi-token prediction layer for faster decoding Apache 2.0 – a permissive licence without territorial restrictions Strong benchmarks: GPQA Diamond 90.4, SWE-bench Verified 78% Inexpensive API: per the Tencent Cloud price list US$0.132 input / US$0.528 output per 1M tokens (TokenHub)
Coding behind GLM-5.2, Claude Opus 4.8 and GPT-5.5 Multimodality (vision) and knowledge cutoff not documented API via Tencent Cloud TokenHub (Chinese provider, no EU region) – for EU companies only self-hosting is advisable
Current
Hy3-FP8
6 July 2026
FP8-quantised variant for efficient self-hosting Reduced memory footprint at near-identical quality
Quantisation can cost quality in edge cases
Current
Older, deprecated and retired versions (1)

Legacy and deprecated: superseded by newer models; for many versions the provider has already announced a shutdown. Retired: no longer available. Dates and successors are in the model lifecycle section.

ModelReleaseEUStrengthsWeaknessesStatus
Hy3-Preview
23 April 2026
Technical predecessor of the full release
Licence with territorial restrictions (no EU use) Superseded by the Apache 2.0 full release; no longer listed in the TokenHub price list
Legacy

Use Cases

Typical applications for this model

Self-hosting in your own cloud
Cost-efficient reasoning workloads
Agentic workflows
Research and analysis (strong on BrowseComp)
Open-source research

Technical Details

API, features and capabilities

API & Availability
Availability Public (Tencent Cloud TokenHub, OpenRouter); weights on Hugging Face and ModelScope
Features & Capabilities
Tool Use Function Calling Structured Output Reasoning Mode
Training & Knowledge
Knowledge Cutoff not documented
Fine-Tuning Available (LoRA, Full)
Language Support
Best Quality Chinese, English
Supported multilingual
Best quality in Chinese and English

Hosting & Compliance

GDPR-compliant hosting options and licensing

GDPR-Compliant Hosting Options
Self-Hosted
Own infrastructure / own cloud (e.g. STACKIT, Azure, AWS)
Open weights under Apache 2.0 – full data sovereignty when self-operated
License & Hosting
License Apache 2.0 (Hy3 full release since 6 July 2026, Hy4-Preview since 28 August 2026; the April 2026 Hy3 preview licence excluded the EU)
Security Filters Customisable (open weights)
On-Premise

Benchmarks

Performance comparison with standardized tests

Internal blind evaluation, 203 engineering tasks (Hy4-Preview)
2.99
Internal blind evaluation, 203 engineering tasks (GLM-5.3, for comparison)
2.92
Internal blind evaluation, 203 engineering tasks (Kimi K3, for comparison)
2.94
GPQA Diamond
90.4
SWE-bench Verified
78
SWE-bench Pro
57.9
BrowseComp
84.2

Tencent Hy3 is the third generation of the Hunyuan model family and has been available as a full release since July 6, 2026 – for the first time under a genuine Apache 2.0 licence without territorial restrictions. That is the real news for European companies: the April preview’s licence still explicitly excluded use in the EU, the UK and South Korea; the final release lifts this restriction. With Hy4-Preview (28 August 2026), a considerably larger open-weights preview (770B vs. 295B total parameters) is also available under Apache 2.0; Hy3 remains the stable reference for production self-hosting. This overview reflects the state as of October 2026.

Architecture: Efficiency over Size

Hy3 is a mixture-of-experts model with 295 billion total and only 21 billion active parameters (192 experts, 8 active per token). Add a 256K token context window, an additional multi-token prediction layer for faster decoding, and a hybrid fast/slow thinking design that toggles between quick replies and deep reasoning. Besides the standard version, an FP8-quantised variant is available for efficient self-hosting.

The positioning is clear: Hy3 is the efficiency play among Chinese open-source models – considerably smaller than DeepSeek V4-Pro (1.6 trillion parameters) and GLM-5.3 (around 753 billion per the model card), with competitive results in most disciplines except coding.

Benchmarks

According to Tencent and independent evaluations, Hy3 achieves, among others, 90.4 on GPQA Diamond, 78% on SWE-bench Verified, 57.9% on SWE-bench Pro and 84.2 on BrowseComp. On demanding maths (MathArena Apex: 38.7 vs. 85.4 for GPT-5.5) and in coding, the model trails the Western flagships and GLM-5.2. For the model size – only 21B active parameters – the results are remarkable.

Hy4-Preview: the Next Generation

With Hy4-Preview, Tencent released the successor as an open-weights preview on August 28, 2026 (Hugging Face: tencent/Hy4-preview, plus an FP8 variant tencent/Hy4-preview-FP8). Both are listed under Apache 2.0 per the Hugging Face model card – unlike Hy3, there is no territorial restriction from the outset.

Key specs (vendor-reported):

  • 770B total, 49B active parameters (MoE) – considerably larger than Hy3
  • Context window of 1M tokens per the model card
  • In an internal blind evaluation by Tencent (163 experts, 203 engineering tasks), Hy4-Preview scores 2.99 out of 4.00, narrowly ahead of GLM-5.3 (2.92) and Kimi K3 (2.94)
  • Integrated into Tencent’s own products WorkBuddy, CodeBuddy, Yuanbao and ima from launch
  • API via Tencent Cloud TokenHub: per the price list US$0.834 / 1M input, US$2.501 / 1M output, US$0.042 / 1M cache

Context: this is explicitly a preview, not a final full release. As known issues, Tencent’s model card lists reasoning longer than necessary and a tendency to over-verify its own work. The comparison figures against GLM-5.3 and Kimi K3 come from an internal Tencent evaluation; we have no independent third-party confirmation. For production self-hosting deployments we therefore continue to recommend Hy3 as the established, stable full release, while watching how Hy4 develops.

Availability and Cost

The weights are available on Hugging Face and ModelScope; as an API, Hy3 is offered via Tencent Cloud’s TokenHub (per the price list US$0.132 input / US$0.528 output per 1M tokens; Singapore, Guangzhou and Silicon Valley regions) and via OpenRouter. Within Tencent’s own ecosystem, Hy3 powers the WeChat assistant, Yuanbao and CodeBuddy, among others.

GDPR and Data Sovereignty: Self-Hosting Yes, China API with Caution

For European companies, the same logic applies as with DeepSeek and Qwen: Hy3 is interesting above all as an open-weights model for self-operation. Under Apache 2.0 it can be hosted in your own cloud environment – on STACKIT, Azure or AWS, for example – meaning no data flows to the vendor and full data sovereignty is maintained. The API via Tencent Cloud, by contrast, means processing by a Chinese provider and should be assessed critically from a GDPR perspective for personal or confidential data.

When reviewing the licence, mind the version: only the full release of July 6, 2026 and Hy4-Preview from August 28, 2026 are under Apache 2.0 – anyone still using artifacts of the April 2026 Hy3 preview operates under the old community licence with its EU exclusion.

Hyperscaler catalogues: We found no native offering for Hunyuan on AWS Bedrock (no entry in the Bedrock region table), the Google Gemini Enterprise Agent Platform (formerly Vertex AI; not in the list of managed open models) or Microsoft Foundry. Hunyuan is therefore not listed as a managed model in any of the major hyperscaler catalogues; self-hosting remains the only route to EU data processing that we are aware of.

Positioning Among Chinese Models

ModelSize (active)LicenceDistinctive feature
Tencent Hy4-Preview770B (49B)Apache 2.0Newest model, still preview status
Tencent Hy3295B (21B)Apache 2.0Efficiency, licence without territorial restrictions
DeepSeek V41.6T (Pro, 49B active)MIT1M context, focus on agents and coding
GLM-5.3approx. 753Bbespoke GLM-5.3 licenceCoding focus per Z.ai
Qwen 3.827.8B (dense) to 2.4T (Max)Apache 2.0 (27B), bespoke licences (Flash-Next, Max)Broad range, API via Alibaba Cloud

Integration with CompanyGPT

As an open-weights model, Hy3 – like Llama, Qwen or GptOSS – can be integrated into a self-hosted platform such as CompanyGPT and operated GDPR-compliantly in your own cloud. For most enterprise use cases in the DACH region, we continue to recommend the established frontier models via EU endpoints; Hy3 is an exciting option for cost-efficient, sovereignly hosted workloads.

For individual consulting on the right model strategy, contact innFactory AI Consulting.

Cost estimation for this model

For up-to-date token pricing, model variants and EU availability, see our sister project ai-prices.eu. It helps you compare and estimate the operational cost of leading AI models for your specific use case.

Compare prices on ai-prices.eu

ai-prices.eu is a project by innFactory AI Consulting GmbH and provides transparent cost estimates for leading AI models.

Consultation for this model?

We help you select and integrate the right AI model for your use case.