Hy Image 3.5 Preview
Tencent Hunyuan
- Released
- September 22, 2026
- Data date
- October 7, 2026
- Model type
- Maximum resolution
Hy Image 3.5 Preview focuses on clean text rendering and complex layouts, including small text and mixed text-image compositions. The product announcement lists up to five reference images and 2K output; the separate TokenHub API guide lists up to 4K for that API access path.
Specifications and access
| Specification | Value and source |
|---|---|
| Model type | Proprietary modelSource |
| API model ID | hy-image-v3.5-previewSource |
| Access | Tencent Cloud account, TokenHub API key, and enabled postpaid model requiredSource |
| Release | Preview released on September 22, 2026Source |
| Capabilities | Text-to-image, image-to-image, and multi-turn editing with reference imagesSource |
| Reference images | Up to five reference images in the product announcementSource |
| Product-announcement resolution | Up to 2KSource |
| TokenHub API guide resolution | Up to 4096 × 4096 (4K)Source |
| Maximum resolution | Product announcement: up to 2K; TokenHub API guide: up to 4KSource |
| Native audio | No native audio output documentedSource |
| Generator | hy-image-v3.5-previewSource |
| Input modalities | Text and reference imagesSource |
| Output modalities | ImagesSource |
Measurements without matching peer values
These measurements have no matching peer values under the same test conditions. Their original values and sources remain available here.
AA-Image-T2I overall preference Elo (HunyuanImage 3.5 Preview) (Show measurement, test conditions, and source)
- Source value
- 975
- Score
- 975
- Metric
- Elo from pairwise human preference votes
- Unit
- elo
- Benchmark version
- AA-Image-T2I v2.0, Overall
- Category
- text-to-image
- Direction
- Higher is better
- Sample
- 5498
- Confidence interval
- 9
- Lower confidence bound
- 966
- Upper confidence bound
- 984
- Source type
- independent-evaluator
- Evaluator
- Artificial Analysis AA-Image-T2I v2.0 leaderboard
- Status
- active
- Retrieved at
- 2026-10-07
- Methodology
- Independent evaluation. Overall text-to-image board. Artificial Analysis lists the Tencent model as HunyuanImage 3.5 (Preview). Matching it to profile hy-image-3.5-preview is an identity inference based on the same provider and version label, checked against Tencent’s official Hy Image 3.5 Preview announcement and TokenHub service ID hy-image-v3.5-preview. The board ranks text-prompt image generation by Elo from pairwise human preference votes.
- Context
- The leaderboard labels 5,498 as samples; it does not establish that this is a count of unique prompts. This independent Elo score is not Tencent’s GSB comparative claim.
AA-Image-Editing overall preference Elo (HunyuanImage 3.5 Preview) (Show measurement, test conditions, and source)
- Source value
- 1087
- Score
- 1087
- Metric
- Elo from pairwise human preference votes
- Unit
- elo
- Benchmark version
- AA-Image-Editing v2.0, Overall
- Category
- image-editing
- Direction
- Higher is better
- Sample
- 5907
- Confidence interval
- 9
- Lower confidence bound
- 1078
- Upper confidence bound
- 1096
- Source type
- independent-evaluator
- Evaluator
- Artificial Analysis AA-Image-Editing v2.0 leaderboard
- Status
- active
- Retrieved at
- 2026-10-07
- Methodology
- Independent evaluation. Overall image-editing board. Artificial Analysis lists the Tencent model as HunyuanImage 3.5 (Preview). Matching it to profile hy-image-3.5-preview is an identity inference based on the same provider and version label, checked against Tencent’s official Hy Image 3.5 Preview announcement and TokenHub service ID hy-image-v3.5-preview. The board ranks image editing by Elo from pairwise human preference votes.
- Context
- The leaderboard labels 5,907 as samples; it does not establish that this is a count of unique prompts. This independent Elo score is not Tencent’s GSB comparative claim.
Sources and data date
| Type | Evidence and data date |
|---|---|
| Research status | Research date October 7, 2026. 6 source URLs checked. This documents the inspected sources, not an exhaustive inventory of every publication. |
| Not published | interpretable Tencent GSB measurement and evaluation details Original wording Tencent says that more than 100 professional designers’ blind GSB comparison against HY Image 3.0 showed a comprehensive win-rate improvement of more than 30%. The source does not clarify whether that means relative growth or percentage points, and publishes no exact uplift, absolute win rate, prompts, or case count. Because its numeric basis is ambiguous, this vendor claim is described here rather than entered as a scored benchmark. The separate independent AA image-generation and image-editing Elo results are listed above. |
| Additional source | Tencent Hunyuan Hy Image 3.5 Preview announcement (retrieved October 7, 2026) · Tencent Hunyuan Hy Image 3.5 Preview announcement · Editorial description reviewed October 7, 2026 |
| Additional source | Tencent TokenHub Hy image generation API guide (retrieved October 7, 2026) · Tencent TokenHub Hy image generation API guide · Editorial description reviewed October 7, 2026 |
| Additional source | Artificial Analysis AA-Image-T2I v2.0 leaderboard (retrieved October 7, 2026) |
| Additional source | Artificial Analysis AA-Image-Editing v2.0 leaderboard (retrieved October 7, 2026) |