What is Hunyuan Vision 1.5 Thinking?
Hunyuan Vision 1.5 Thinking is a vision-language model that reasons step by step before it answers. Tencent built it on a hybrid Mamba-Transformer design and aims it at image questions, OCR, charts, locating objects and solving problems from a photo, with better English and smaller-language support than its earlier vision models. It reads images and text and writes text, through an OpenAI-compatible API.
Tencent Cloud's China TokenHub still lists it with no retirement date, but it is no longer Tencent's newest vision model: HY-Vision-2.0-Instruct is newer and answers without a thinking step. The open weights and technical report Tencent promised for Hunyuan-Vision-1.5 have not appeared.
Is Hunyuan Vision 1.5 Thinking still available?
Yes, on Tencent Cloud's China TokenHub, which lists it with no retirement date, but Tencent has newer vision models and never released its weights.
- TokenHub (China)
- Listed as HY-Vision-1.5-Thinking at ¥3 / ¥9 per million tokens, with no shutdown date.
- Older Hunyuan API
- Tencent is moving its original Hunyuan platform to TokenHub; existing buyers keep access, but it takes no new purchases.
- International TokenHub
- Not listed; Tencent's international model list has no HY-Vision models.
- Newer model
- HY-Vision-2.0-Instruct, a faster model without a thinking step, at ¥7.5 / ¥17.5.
Who is Hunyuan Vision 1.5 Thinking best for?
Hunyuan Vision 1.5 Thinking is best for developers on Tencent Cloud in China who need a low-cost reasoning model to read images: OCR, charts, visual grounding and worked answers to photographed questions.
OCR on photos and scans
Tencent names reading text in images among its core tasks, and a 1024 x 1024 image costs only about 1,058 input tokens.
Charts and visual grounding
As a thinking model it reasons before it answers, which suits charts, diagrams and finding objects in a picture.
Study and tutoring apps
It can work out the answer from a photo of a question, a task Tencent names for this model, and its 24K-token output leaves room for worked steps.
Teams on Tencent Cloud in China
It bills by the token in yuan through TokenHub, on the same account as Tencent's other models.
What's included with Hunyuan Vision 1.5 Thinking?
Step-by-step reasoning, an OpenAI-compatible API, images by link or Base64, several images per request and answers in English and smaller languages.
- Thinks before it answers
- It works through a chain of reasoning first, which Tencent aims at charts, grounding and multi-step problems.
- OpenAI-compatible API
- Call it through the Chat Completions format on TokenHub with the model ID hunyuan-t1-vision-20250916.
- Images by link or Base64
- Send public image URLs or Base64 data in JPEG, PNG, GIF or WEBP, ideally up to 10 MB per image.
- Several images per request
- There is no fixed image cap; the 16K-token input limit decides how many fit.
- Multilingual answers
- Tencent tuned it for English and smaller languages, not only Chinese.
How much does Hunyuan Vision 1.5 Thinking cost?
Hunyuan Vision 1.5 Thinking is pay as you go on Tencent Cloud's China TokenHub, at ¥3 per million input tokens and ¥9 per million output tokens, billed every hour after use.
- Billed in Chinese yuan, settled every hour
- 40K-token context, up to 24K output tokens
- Input and output tokens charged separately
Tencent lists this price in Chinese yuan on its China site, shows no US dollar price, and does not offer the model on its international TokenHub. Tencent's monthly Token Plan subscriptions do not cover image input, so pay as you go is the only way to use it. Bills arrive two to three hours after use. The newer HY-Vision-2.0-Instruct costs ¥7.5 per million input tokens and ¥17.5 per million output tokens. Full pricing on cloud.tencent.com. We checked these prices on .
How do readers rate Hunyuan Vision 1.5 Thinking?
No one has rated Hunyuan Vision 1.5 Thinking yet. Be the first.
Sign in to rate. One rating per account.
Hunyuan Vision 1.5 Thinking pros and cons
It is cheap and reasons through charts and photographed problems, but its context is small, it is sold only on Tencent's China cloud, and the promised open weights have not shipped.
Pros
- Costs ¥3 per million input tokens and ¥9 per million output, well below the ¥7.5 / ¥17.5 of Tencent's newer HY-Vision-2.0-Instruct.
- Its 24K-token output limit is larger than its 16K input limit, which leaves room for long worked answers to charts and photographed problems.
- Uses the OpenAI Chat Completions format, so existing OpenAI client code works with a new base URL and model ID.
Cons
- The context is small for a vision model: 40K tokens in total and 16K of input, against 128K for GLM-4.6V and 262K natively for Qwen3.6-27B.
- Tencent sells it only on its China cloud, priced in yuan; the international TokenHub does not list it.
- The open weights, technical report and "thinking on images" mode Tencent promised for Hunyuan-Vision-1.5 have not shipped.
- Tencent documents it for images only and sends video work to a separate model, HY-Vision-Video.
What are the best Hunyuan Vision 1.5 Thinking alternatives?
Three tools on Recaply do Hunyuan Vision 1.5 Thinking's job. All three are open source, so you can run them on your own server.
GLM-4.6VOpen SourceChoose it when you want a vision model billed in US dollars with a 128K context and video input; it costs $0.3 / $0.9 per million tokens, and its MIT-licensed weights are free to download.
Step3-VL-10BOpen SourceChoose it when you want a reasoning vision model on your own GPU: its 10B weights are Apache 2.0, and it reasons over up to 64K tokens in its standard mode.
Qwen3.6-27BOpen SourceChoose it when you need long context and video input: Alibaba's Model Studio serves it with up to 256K tokens at $0.60 / $3.60 per million tokens, and its Apache 2.0 weights are open.
Questions about Hunyuan Vision 1.5 Thinking
What is Hunyuan Vision 1.5 Thinking?
Hunyuan Vision 1.5 Thinking is Tencent's reasoning vision model for reading images, charts and photographed questions, sold by the token on Tencent Cloud's China TokenHub for developers building OCR and tutoring apps.
Who is Hunyuan Vision 1.5 Thinking for?
Hunyuan Vision 1.5 Thinking is best for developers on Tencent Cloud in China who need a low-cost reasoning model to read images: OCR, charts, visual grounding and worked answers to photographed questions.
Is Hunyuan Vision 1.5 Thinking free?
No. Hunyuan Vision 1.5 Thinking is a paid tool, from ¥3 / ¥9 per million input / output tokens.
How much does Hunyuan Vision 1.5 Thinking cost?
Hunyuan Vision 1.5 Thinking's paid plans are Pay as you go at ¥3 / ¥9 per million input / output tokens.
Is there an open source alternative to Hunyuan Vision 1.5 Thinking?
Yes. GLM-4.6V, Step3-VL-10B and Qwen3.6-27B are open source tools that do the same job, so you can run your own.
What are the best Hunyuan Vision 1.5 Thinking alternatives?
The closest alternatives on Recaply are GLM-4.6V, Step3-VL-10B and Qwen3.6-27B.
Can I download Hunyuan Vision 1.5 Thinking and run it myself?
No. Tencent said it would release Hunyuan-Vision-1.5 checkpoints in two sizes and a technical report, but that checklist on its GitHub page is still unticked. For now the model is only available through Tencent Cloud's API.
What are Hunyuan Vision 1.5 Thinking's context limits?
Hunyuan Vision 1.5 Thinking has a 40K-token context window on TokenHub, takes up to 16K input tokens and returns up to 24K output tokens, so its output limit is larger than its input limit and leaves room for long step-by-step answers.
How does Hunyuan Vision 1.5 Thinking compare with HY-Vision-2.0-Instruct?
HY-Vision-2.0-Instruct is Tencent's newer vision model and costs more, at ¥7.5 and ¥17.5 per million input and output tokens against ¥3 and ¥9. Tencent says 2.0 improves perception, OCR, STEM and chart reading and answers quickly, while 1.5 Thinking reasons before it answers.
How many images can I send to Hunyuan Vision 1.5 Thinking at once?
There is no fixed limit; the context window decides. Tencent estimates about 1,058 tokens for a 1024 x 1024 image, so a 16K-token input fits roughly 15 such images, and it warns that answers get worse when a request carries too many.
Our verdict
Hunyuan Vision 1.5 Thinking makes sense only if you already build on Tencent Cloud in China and want cheap image reasoning.
It costs less than Tencent's newer HY-Vision-2.0-Instruct and thinks before it answers, but its 40K context is small and the open weights Tencent promised have not shipped.
- Skip it for
- Teams outside Tencent's China cloud, anyone who needs video or long documents, and anyone who wants to self-host.
- Check first
- Run the same images through HY-Vision-2.0-Instruct first; it is newer and may read text and charts better without the thinking step.
Nomos 1
Gemini 3.1 Deep Think
Brave Ocelot
Hydrogen-1.0
Biohub ESM
MiroThinker