Last verified
WECHAT-SCALE LABFOUNDED 1998SHENZHEN · CHINATENCENT CLOUDWECHAT PARENT

Hunyuan API Pricing

The Hunyuan (混元) model family from Tencent — parent of WeChat, QQ, and the world's largest games business. Served through Tencent Cloud and the Yuanbao assistant inside WeChat, Hunyuan reaches over a billion users through China's dominant super-app.

Production models
9
8 GA · 1 preview
Founded
1998
Tencent · Shenzhen
Newest model
AUG 28
Hunyuan Hy4 Preview launch
WeChat reach
1.4B
MAU · Yuanbao access
Context window
1M
Hunyuan Hy4 Preview
Top of range
6.3x
Hy4 Preview input vs Hy3
§ 01 / LINEUP

The full roster.

Side-by-side →
FAST · LIGHTWEIGHT

Hunyuan A13B

Hunyuan MoE 13B-active model.

Input
$0.0744/M
Output
$0.298/M
224K ctx · text-only
FRONTIER · REASONING

Hunyuan Hy3 LAUNCHED JUL 2026

Tencent's newest GA Hunyuan flagship, released 2026-07-06.

Input
$0.132/M
Output
$0.528/M
256K ctx · text-only
BALANCED · MID-TIER

Tencent HY Vision 1.5 Instruct

Older Tencent vision-language baseline, distinct from TurboS Vision and T1 Vision.

Input
$0.446/M
Output
$1.34/M
24K ctx · vision
BALANCED · MID-TIER

Hunyuan T1 Vision

Tencent's multimodal deep-reasoning sibling of T1.

Input
$0.446/M
Output
$1.34/M
28K ctx · vision
FAST · LIGHTWEIGHT

Hunyuan Translation

Translation-specialised.

Input
$0.178/M
Output
$0.536/M
documented elsewhere ctx · text-only
FAST · LIGHTWEIGHT

Hunyuan Translation Lite

Light translation variant.

Input
$0.149/M
Output
$0.446/M
documented elsewhere ctx · text-only
BALANCED · MID-TIER

Hunyuan TurboS Vision

Tencent's fast vision-language flagship.

Input
$0.446/M
Output
$1.34/M
32K ctx · vision
BALANCED · MID-TIER

Hunyuan TurboS Vision Video

Video-capable variant of TurboS Vision for clip description and video Q&A.

Input
$0.446/M
Output
$1.34/M
24K ctx · vision
EMBEDDING

Hunyuan Embedding

Tencent's text embedding endpoint.

Input
$0.104/M
Output
$0.104/M
documented elsewhere ctx · text-only
PREVIEW · EARLY ACCESS

Hunyuan Hy4 Preview LAUNCHED AUG 2026

Tencent's newest Hunyuan flagship, open-sourced 2026-08-28 - 770B total / 49B active MoE, context window over 1M tokens (vendor states 1M window, 960k max input, 64k max output).

Input
$0.834/M
Output
$2.50/M
1M ctx · text-only
DEPRECATED · RETIRING

Hunyuan 2.0 Think (HYThink)

DEPRECATED.

Input
$0.591/M
Output
$2.37/M
128K ctx · text-only
DEPRECATED · RETIRING

Hunyuan 2.0 Instruct

DEPRECATED.

Input
$0.473/M
Output
$1.18/M
128K ctx · text-only
DEPRECATED · RETIRING

Hunyuan Large Role

Role-play specialised Hunyuan.

Input
$0.357/M
Output
$1.43/M
28K ctx · text-only
LEGACY · STILL SUPPORTED

Hunyuan T1

Tencent Hunyuan reasoning-tier model.

Input
$0.149/M
Output
$0.595/M
documented elsewhere ctx · text-only
LEGACY · STILL SUPPORTED

Hunyuan TurboS

Tencent's high-speed/budget tier.

Input
$0.119/M
Output
$0.298/M
documented elsewhere ctx · text-only
LEGACY · STILL SUPPORTED

Hunyuan Lite

FREE tier.

Input
Free/M
Output
Free/M
documented elsewhere ctx · text-only
§ 02 / SHELF

All side-by-side.

Methodology →
Model Input /M Output /M Cached Context Max output Vision Tools Tier
Hunyuan A13B $0.0744 $0.298 224K Light
Hunyuan Hy3 FLAGSHIP $0.132 $0.528 $0.033−75% 256K Frontier
Tencent HY Vision 1.5 Instruct $0.446 $1.34 24K Mid
Hunyuan T1 Vision $0.446 $1.34 28K Mid
Hunyuan Translation $0.178 $0.536 documented elsewhere Light
Hunyuan Translation Lite $0.149 $0.446 documented elsewhere Light
Hunyuan TurboS Vision $0.446 $1.34 32K Mid
Hunyuan TurboS Vision Video $0.446 $1.34 24K Mid
Hunyuan Embedding $0.104 $0.104 documented elsewhere Active
Hunyuan Hy4 Preview PREVIEW $0.834 $2.50 $0.042−95% 1M Preview
Hunyuan 2.0 Think (HYThink) DEPRECATED $0.591 $2.37 128K Deprecated
Hunyuan 2.0 Instruct DEPRECATED $0.473 $1.18 128K Deprecated
Hunyuan Large Role DEPRECATED $0.357 $1.43 28K Deprecated
Hunyuan T1 LEGACY $0.149 $0.595 documented elsewhere Legacy
Hunyuan TurboS LEGACY $0.119 $0.298 documented elsewhere Legacy
Hunyuan Lite LEGACY Free Free documented elsewhere Legacy
§ 03 / PRICE CURVE

Pricing across the lineup.

How Tencent (Hunyuan) priced 7 models · FEB 25 → AUG 26.

oldest → newest →
$5$3.75$2.5$1.25 $0.595 $0.149 $0.298 $0.119 $1.34 $0.446 $1.18 $0.473 $2.37 $0.591 $0.528 $0.132 $2.5 $0.834 Hunyuan T1 FEB 25 Hunyuan TurboS MAR 25 Hunyuan TurboS … MAY 25 Hunyuan 2.0 Ins… NOV 25 Hunyuan 2.0 Thi… NOV 25 Hunyuan Hy3 JUL 26 Hunyuan Hy4 Pre… AUG 26
Input · newest $0.834/M
Output · newest $2.5/M
Each point is a model at its listed $/M price.

The Hunyuan story so far.

Releases, pricing, and WeChat integration from Tencent's AI line. Sourced from Tencent Cloud, Tencent press, and Wikipedia — verified at publication.

AUG 31 · 2026
The original Hunyuan platform stops taking new customers — capabilities move to TokenHub, and T1, TurboS and Lite drop off the price list entirely
LIFECYCLE
AUG 28 · 2026
Hy4 Preview released and open-sourced — 770B total / 49B active, 1M context, listed at 6.3x the Hy3 input rate
RELEASE
JUL 06 · 2026
Hunyuan Hy3 released — 295B/21B hybrid fast/slow-thinking MoE, 256K context, Apache 2.0 open weights
RELEASE
APR · 2026
Hunyuan 3.0 planned — Tencent says it will build a WeChat AI agent on the new generation
RELEASE
MAR 13 · 2026
Tencent raised Hunyuan 2.0 list prices ~450% — Think and Instruct repriced upward vs prior rates
PRICING
NOV 09 · 2025
Hunyuan 2.0 released — Think + Instruct tiers, 128K context
RELEASE
MAY 21 · 2025
Hunyuan model matrix upgraded — TurboS + T1 + new T1-Vision and Hunyuan Voice
RELEASE
MAR · 2025
Hunyuan TurboS released — first industry-deployed large-scale Hybrid-Transformer-Mamba MoE; T1 reasoning model follows
RELEASE
MAY · 2024
Yuanbao (元宝) consumer assistant launched — later embedded directly into WeChat
RELEASE
SEP · 2023
Hunyuan large model line first announced by Tencent
CORPORATE
§ 04 / ACCESS

Where to get it.

Methodology →
§ 04 / BEST FOR

Which Tencent (Hunyuan) for what.

More scenarios →
If you need the biggest model and 1M context
Hunyuan Hy4 Preview
Profile →
If you want steady production traffic
Hunyuan Hy3
Profile →
If you want the cheapest tokens, or open weights
Hunyuan A13B
Profile →
If you process images
Tencent HY Vision 1.5 Instruct
Profile →
If you process video
Hunyuan TurboS Vision Video
Profile →
If you need machine translation
Hunyuan Translation
Profile →
If you need China data residency
Tencent Cloud (China)
Tencent Cloud →
§ 06 / BACKGROUND

The company behind it.

www.tencent.com →

Tencent was founded in 1998 in Shenzhen by Pony Ma (Ma Huateng) and co-founders. It is China's largest technology company — owner of WeChat / Weixin, QQ, and the world's biggest games business — and is publicly listed (HKEX: 0700), reporting roughly $109 billion in 2025 revenue.

Tencent's AI line is the Hunyuan (混元) family, served through Tencent Cloud and the Yuanbao (元宝) consumer assistant. Yuanbao launched in May 2024 and was later embedded directly inside WeChat, giving Hunyuan a distribution path to over a billion users — a structural advantage no Western lab can match in China.

The current lineup is headed by Hy4 Preview (770B total / 49B active, 1M context, open-sourced August 2026) sitting above Hy3, the generally available 295B/21B flagship. Below them sit A13B — the cheapest row and the self-hostable one — plus vision, video, translation and embedding specialists. The older platform's models (Hunyuan 2.0 Think and Instruct, T1, TurboS, Lite) are off the price list: as of August 2026 Tencent has moved model serving to TokenHub, and the original platform adds no new models and takes no new customers, though existing ones keep running.

Pricing needs care here, because Tencent publishes two separate lists. Tencent Cloud International prices Hy4 Preview and Hy3 directly in USD, while the CN-region console prices them in yuan — and the two are not the same figure in two currencies. Converting the CNY card overstates the International rate by about 7%, so this site reads the USD list where one exists and converts only where it does not. Tencent also raised Hunyuan 2.0 list prices roughly 450% on March 13, 2026, and priced Hy4 Preview at 6.3x the Hy3 input rate — two upward moves in a market racing to the bottom, which we flag honestly. A Hunyuan 3.0 generation is planned for 2026.

The trade-offs: Tencent Cloud is a China-hosted service, so the direct API stores data in mainland China, and US export-control / procurement questions apply. Versus ByteDance's Doubao it leads on super-app reach via WeChat; versus Alibaba's Qwen and DeepSeek it open-sources less and recently moved prices up rather than down.

§ 07 / COMPETITORS

Other frontier labs.

All providers →

Frequently asked.

Practical questions about Hunyuan pricing, deployment, and data residency.

Q · 01 Which Hunyuan model should I start with? +
For most production work: Hunyuan Hy3 ($0.132/$0.528), the generally available flagship. When you need 1M context or the largest model Tencent sells: Hy4 Preview ($0.834/$2.501), though it carries preview status. For the cheapest tokens or self-hosting: A13B ($0.0744/$0.2976). Note that the models this answer used to recommend — Hunyuan 2.0 Think and Instruct, TurboS, and the free Lite tier — are no longer on Tencent's price list. See the use case picker above.
Q · 02 Why did Hunyuan prices go up? +
Twice, and for different reasons. On March 13, 2026 Tencent raised Hunyuan 2.0 list prices by roughly 450% — an unusual upward move while most labs cut. Then on August 28, 2026 it launched Hy4 Preview at 6.3x the Hy3 input rate, which is a new tier above rather than a reprice: Hy3 stays on the list unchanged. On sourcing: rates for Hy4 Preview and Hy3 come from Tencent's own International USD list; the remaining CN-only rows are converted from CNY at ¥6.7209/USD, and the older Hunyuan 2.0 rows quoted here are the ≤32K tier.
Q · 03 Where is my data stored? +
It depends which endpoint you call, and this is worth getting right. The CN-region Tencent Cloud endpoint is China-hosted, so data is processed and stored in mainland China. But TokenHub International publishes non-China endpoints — Singapore (tokenhub-intl.tencentcloudmaas.com) and Silicon Valley (tokenhub-us.tencentcloudmaas.com) — and its prices are the USD figures on this page. The Yuanbao assistant lives inside WeChat under Chinese data rules regardless. US export-control and procurement questions still apply for some buyers; self-hosting the open weights remains the strongest form of data control.
Q · 04 Are Hunyuan models open-weight? +
Increasingly, yes — and that changed at the top of the range. Hy3 shipped with Apache 2.0 weights on Hugging Face and ModelScope, and Tencent's Hy4 Preview announcement is headlined "Releases and Open-Sources", shipping a checkpoint, an FP8 variant and serving recipes; that release does not name a licence, so we do not quote one. A13B and Hunyuan-Large remain open under Tencent's community licence. The older API-only tiers this answer used to name are off the price list.
Q · 05 How does Hunyuan reach so many users? +
Through WeChat. Tencent embedded the Yuanbao assistant directly into the super-app, so over a billion WeChat users can use Hunyuan without installing anything. That distribution — not raw benchmarks — is Tencent's biggest AI advantage.
Q · 06 How does Tencent compare to other Chinese labs? +
Versus ByteDance's Doubao, Tencent leads on WeChat-embedded reach but trails on standalone-app usage. Versus Qwen and DeepSeek it open-sources less and recently moved prices up — so for pure cost or open weights, those peers are stronger.