Tencent Hunyuan API pricing: Hy4, Hy4 Flash and the half-price deal
Updated 2026-11-20. Official Tencent Cloud prices, LMArena ranks.
Tencent's Hy4 costs $0.834 per 1M input tokens and $2.501 output, and Hy4 Flash $0.07 and $0.281, per Tencent Cloud's pricing page. Tencent bills Hy4 at half price until Dec 31, 2026 and Hy4 Flash until Nov 30 (Beijing time). Both have 1M context, and the weights are open on Hugging Face.
Tencent's Hunyuan API prices Hy4, its flagship, at $0.834 per million input tokens and $2.501 per million output tokens, and Hy4 Flash at $0.07 and $0.281, per Tencent Cloud's TokenHub pricing page. Both are half price for now: Hy4 until Dec 31, 2026 and Hy4 Flash until Nov 30, Beijing time.
The other news: as of Nov 20, 2026 the full Hy4 weights are public on Hugging Face, following Hy4 Flash on Nov 17. So you can pay Tencent per token or run the same model on your own GPUs.
Tencent Hunyuan API prices per 1M tokens
| Model | Input | Cached | Output | Context |
|---|---|---|---|---|
| Hunyuan Hy4 | $0.834 | $0.042 | $2.501 | 1M |
| Hunyuan Hy4 Flash | $0.07 | $0.003 | $0.281 | 1M |
| Hunyuan Hy3 | $0.132 | $0.033 | $0.528 | 256K |
| Hunyuan Hy4 Preview (being discontinued) | $0.834 | $0.042 | $2.501 | 1M |
USD per 1M tokens, Tencent Cloud TokenHub list price (Singapore tab), checked 2026-11-20, from Tencent's model pricing page. API names: hy4, hy4-flash, hy3, hy4-preview. No discounts applied.
Every Hy4 model has a 1M-token window, with up to 960K tokens of input and 64K of output per request, per Tencent's model list. Hy3 tops out at 256K. Tencent lists no higher price for long prompts, so a 900K-token prompt costs the same per token as a short one. Cache hits are 95% off input on Hy4 and 96% on Hy4 Flash, and there's no separate cache-write fee.
How long is Hy4 half price?
| Model | Now, in / out | List, in / out | Promo ends |
|---|---|---|---|
| Hunyuan Hy4 | $0.417 / $1.25 | $0.834 / $2.501 | Dec 31, 10:59 AM ET |
| Hunyuan Hy4 Flash | $0.035 / $0.141 | $0.07 / $0.281 | Nov 29, 11:00 PM ET |
Tencent bills usage at 50% of list until 23:59:59 Dec 31, 2026 (Hy4) and 12:00 Nov 30, 2026 (Hy4 Flash), Beijing time, per its pricing page. "Now" is our math: half the list price, rounded to 3 decimals. Usage after the cutoff bills at list.
The Hy4 discount runs through the end of the year, so until New Year Hy4 costs close to what DeepSeek V4.1 Flash costs at list. The Hy4 Flash deal is much shorter and ends at 11 PM ET on Nov 29. Tencent says only usage during the window gets the discount, and it doesn't say whether either promo will be extended.
Which Hunyuan model should you use?
- Hy4 for agent and coding work. It ranks #17 of 45 on LMArena's Agent Arena and #22 of 58 on Code Arena among the models we price. It reads images too.
- Hy4 Flash for cheap bulk jobs. At $0.07 in and $0.281 out it's about 10x cheaper than Hy4 blended, with the same 1M window and image input. LMArena doesn't rank it yet, so test it on your own prompts.
- Hy3 only if you're already on it. It costs more than Hy4 Flash ($0.132 / $0.528), has a quarter of the context and ranks #41 of 58 on Code Arena.
- Hy4 preview: move off it. Tencent marks it "To Be Discontinued". It costs the same as Hy4, doesn't list image input, and Tencent's promo notice names Hy4, not the preview.
Is there a free tier?
Sort of. New TokenHub accounts can claim 1M free tokens per language model, valid for a year, once per model, until Dec 31, 2026, per Tencent's free trial page. The catch: Tencent freezes $30 of your account balance for each Hunyuan model you claim or turn on, and refunds it only after you delete that model's endpoints.
Also note that a request you cancel halfway still bills the tokens already generated, per Tencent's billing rules.
Hy4 vs DeepSeek, Qwen, GLM and Kimi on price
| Model | In / out | Code Arena | Agent Arena |
|---|---|---|---|
| Hunyuan Hy4 | $0.834 / $2.501 | #22 of 58 | #17 of 45 |
| Hunyuan Hy4 Flash | $0.07 / $0.281 | not ranked | not ranked |
| DeepSeek V4.1 Flash | $0.30 / $1.20 | #19 of 58 | #24 of 45 |
| DeepSeek V4 Pro | $1.32 / $3.96 | #46 of 58 | not ranked |
| Qwen4-Max | $1.50 / $4.50 | #9 of 58 | #14 of 45 |
| GLM-5.3 | $1.40 / $4.40 | #27 of 58 | #27 of 45 |
| Kimi K3.1 Pro | $2 / $10 | #15 of 58 | #18 of 45 |
USD per 1M tokens at list price from our API price data, checked 2026-11-20. Ranks are among the priced models in the LMArena leaderboard dataset (CC BY 4.0) as of 2026-11-20.
At list price, Hy4 undercuts DeepSeek V4 Pro, Qwen4-Max, GLM-5.3 and Kimi K3.1 Pro. Only DeepSeek V4.1 Flash (and Tencent's own Hy4 Flash) cost less, and DeepSeek V4.1 Flash also ranks a little higher on Code Arena. Hy4's edge is agents, where it ranks #17 of 45 to DeepSeek V4.1 Flash's #24 of 45. With the promo, Hy4 drops to about $0.417 / $1.25, a blended $0.63 per 1M against DeepSeek V4.1 Flash's $0.52 (our math, 3 parts input to 1 output).
Head-to-heads: Hy4 vs DeepSeek V4 Pro, Hy4 vs Qwen4-Max, Hy4 vs GLM-5.3, Hy4 vs Kimi K3.1 Pro and Hy4 Flash vs DeepSeek V4.1 Flash.
What a month costs
List price, no discounts, our math. Medium is 20M input and 5M output tokens a month; heavy is 200M and 50M.
| Model | Medium month | Heavy month |
|---|---|---|
| Hunyuan Hy4 Flash | $2.81 | $28.05 |
| Hunyuan Hy3 | $5.28 | $52.80 |
| Hunyuan Hy4 | $29.18 | $292 |
At half price, a medium Hy4 month is $14.59 instead of $29.18 (our estimate, before caching).
Can you run Hy4 yourself?
Yes, if you have a GPU server. Hy4 is a mixture-of-experts model with 770B total parameters and 49B active per token, per its model card. The FP8 weights are about 786 GB and the full-precision ones about 1.5 TB, and Tencent's vLLM and SGLang examples spread Hy4-FP8 across 8 GPUs. This isn't a laptop model.
Hy4 Flash is smaller, 220B total and 12B active per its model card, but still needs a very large Mac or a multi-GPU box. Our Hy4 Flash hardware requirements page has the memory math.
Both use the Tencent Hy License, which is based on Apache 2.0 with one big addition: a company selling API access to the model needs a separate license from Tencent if it makes more than $50 million a year or serves a product with more than 100 million monthly users. Internal use is exempt.
What we know and what we don't
| Confirmed by Tencent | Not published |
|---|---|
| Hy4 and Hy4 Flash list prices and both 50% promo end dates | The date Hy4 preview stops serving requests |
| 1M context, 960K max input, 64K max output on the Hy4 family | A long-prompt surcharge (none is listed) or default rate limits for Hunyuan text models |
| 1M free tokens per language model for new accounts, through Dec 31, 2026 | Whether the Hy4 promo gets extended into 2027 |
Our take
Through December, Hy4 is one of the better deals in Chinese AI APIs: it ranks #17 of 45 on Agent Arena and costs $0.417 per million input tokens. After Dec 31 it goes back to $0.834 / $2.501 and the gap to DeepSeek V4.1 Flash widens. Our pick: DeepSeek V4.1 Flash for cheap coding, Hy4 for agent work. If you're on Hy4 preview, switch the model name to hy4 now: same price, and it isn't being retired.
Sources
- Tencent Cloud TokenHub model pricing (official)
- Tencent Cloud TokenHub model list (official)
- TokenHub new user free trial (official)
- TokenHub billing mode (official)
- Hy4 model card, Hugging Face (Tencent)
- Hy4 Flash model card, Hugging Face (Tencent)
- Tencent Hy License 1.0
- LMArena leaderboard dataset (CC BY 4.0)
FAQ
How much does the Hy4 API cost?
$0.834 per 1M input tokens and $2.501 per 1M output tokens at list, with cache hits at $0.042. Tencent bills it at 50% of that until Dec 31, 2026, Beijing time.
How much does Hy4 Flash cost?
$0.07 per 1M input tokens and $0.281 output at list. It's half price until noon on Nov 30, 2026, Beijing time (11 PM ET on Nov 29).
Is Hy4 open source?
The weights are open. Tencent released Hy4, Hy4-FP8 and Hy4 base on Hugging Face under the Tencent Hy License, which requires a separate license only for companies selling model access that have over $50 million in yearly revenue or over 100 million monthly users.
Is the Tencent Hunyuan API free?
Not ongoing. New TokenHub accounts can claim 1M free tokens per language model, valid for a year, until Dec 31, 2026, and Tencent freezes $30 of balance per Hunyuan model while it's enabled.
What is Hy4's context window?
1M tokens, with up to 960K of input and 64K of output per request, per Tencent's model list. Tencent lists no extra charge for long prompts.
Is Hy4 preview being shut down?
Yes. Tencent marks hy4-preview "To Be Discontinued" on its pricing page but hasn't given a date. Hy4 costs the same, so switch the model name to hy4.
Related
- Tencent Hunyuan API prices, every model
- Hunyuan Hy4 API pricing
- Hunyuan Hy4 Flash API pricing
- Best Chinese LLM APIs
- Hy4 vs DeepSeek V4 Pro
- Hy4 Flash vs DeepSeek V4.1 Flash
- Hy4 Flash hardware requirements
- Cheapest LLM API
- GLM API pricing explained
- Kimi API pricing explained
- All AI API prices
More guides
- OpenAI vs Anthropic API pricing
- Qwen API pricing tiers explained
- StepFun API pricing explained
- ByteDance Seed API pricing explained
- DeepSeek Search and Fetch API pricing: what changes Oct 31
- MiniMax API pricing explained
Not sure a plan beats the API for you? Try the plan vs API cost calculator.