AlibabaVideo model
Wan 2.7
In image to video, Wan 2.7 is in place 8 of 21, with text only in place 14. A 5 s clip costs about $0.75.
1,428Arena ratinghigher is better
1,337Arena ratinghigher is better
$0.75$0.15 per secondlower is better
15slongest single cliphigher is better
30out of 100same as last month
Summary
At $0.15 per second of video it is fairly expensive; a typical video model costs $0.12.
New in Wan 2.7 are text-to-video, image-to-video, reference-to-video, and editing in one hosted Alibaba Cloud family. Its variants synchronize audio, support multi-shot storytelling, and output 720p or 1080p up to 15 seconds.
For image-to-video, Alibaba distinguishes first-frame, first-and-last-frame, and audio-driven workflows. The reference variant keeps multiple characters and reference media together; the weights remain proprietary and access runs through the API.
Wan 2.7 benchmarks
Image to video arena
How people rate generated videos in blind comparisons, Elo points · higher is better
20 of 21 models
as of September 22, 2026 · Source: LMArena
Text to video arena
How people rate generated videos in blind comparisons, Elo points · higher is better
20 of 23 models
as of September 22, 2026 · Source: LMArena
Show all 131 individual results from 20 sources
Page 1 of 11
Published benchmarks
LMArena Text to Video
Elo ratings summarize pairwise comparisons within this leaderboard. They are not percentages. The scale does not express absolute quality or linear gaps between models.
higher is betterAs of November 5, 2026Scale: 0 Elo - 1,516 Elo
Methodology and context
- Metric
- elo-rating
- Scale
- No upper bound is documented. The chart spans the recorded values and zero.
- Participants
- 23
- Sample
- 27,626
LMArena Image to Video
Elo ratings summarize pairwise comparisons within this leaderboard. They are not percentages. The scale does not express absolute quality or linear gaps between models.
higher is betterAs of November 5, 2026Scale: 0 Elo - 1,495 Elo
Methodology and context
- Metric
- elo-rating
- Scale
- No upper bound is documented. The chart spans the recorded values and zero.
- Participants
- 21
- Sample
- 85,521
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.
VVA-Bench attack success rate (no-defense baseline) (Show measurement, test conditions, and source)
- Source value
- 100
- Score
- 100
- Metric
- attack success rate (ASR)
- Unit
- %
- Benchmark version
- VVA-Bench Table 5, no-defense baseline
- Category
- image-to-video-safety
- Direction
- Lower is better
- Sample
- 452
- Source type
- independent-evaluator
- Evaluator
- VVA-Bench, VPA-Guard paper Table 5
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. Image-to-video safety benchmark; baseline with no defense; exact Table 5 model column wan_27; 452 unsafe prompts; 45 safe-control prompts are separate. Output resolution not reported.
- Context
- ASR is attack success rate; the separate 45 safe controls are not included in this denominator.
AA Video Arena Elo (Show measurement, test conditions, and source)
- Source value
- 1033
- Score
- 1033
- Metric
- pairwise human-preference Elo
- Unit
- elo
- Benchmark version
- AA-Video-I2V v1.1
- Category
- image-to-video
- Direction
- Higher is better
- Sample
- 2306
- Confidence interval
- 19
- Lower confidence bound
- 1014
- Upper confidence bound
- 1052
- Source type
- independent-evaluator
- Evaluator
- Artificial Analysis AA-Video-I2V v1.1
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Independent evaluation. Image-to-video with audio; exact source row Wan2.7.
- Context
- The generic Wan 2.7 profile is not assigned the separate T2V row Wan2.7-260612. The listed I2V comparison includes audio.
VGI-Bench visual-intelligence score (Show measurement, test conditions, and source)
- Source value
- 35.7
- Score
- 35.7
- Metric
- Overall final score
- Unit
- No unit provided
- Benchmark version
- arXiv:2608.19583v3 (Table 2)
- Category
- image-to-video-reasoning
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- VGI-Bench, arXiv:2608.19583v3
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Independent evaluation. Image-to-video visual-reasoning tasks; text prompt plus input image; 1280×720, 5 or 10 seconds at 30 fps. The common paper protocol evaluates half of each task's instances using a fixed random seed.
- Context
- The tables do not specify a score unit or per-model sample count. No percentage interpretation or denominator is inferred.
VGI-Bench visual-intelligence score (Show measurement, test conditions, and source)
- Source value
- 33.9
- Score
- 33.9
- Metric
- Visual Organization domain average
- Unit
- No unit provided
- Benchmark version
- arXiv:2608.19583v3 (Table 2)
- Category
- visual-organization
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- VGI-Bench, arXiv:2608.19583v3
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Independent evaluation. Image-to-video visual-reasoning tasks; text prompt plus input image; 1280×720, 5 or 10 seconds at 30 fps. The common paper protocol evaluates half of each task's instances using a fixed random seed.
- Context
- The tables do not specify a score unit or per-model sample count. No percentage interpretation or denominator is inferred.
VGI-Bench visual-intelligence score (Show measurement, test conditions, and source)
- Source value
- 36.8
- Score
- 36.8
- Metric
- Spatiotemporal Dynamics domain average
- Unit
- No unit provided
- Benchmark version
- arXiv:2608.19583v3 (Table 2)
- Category
- spatiotemporal-dynamics
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- VGI-Bench, arXiv:2608.19583v3
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Independent evaluation. Image-to-video visual-reasoning tasks; text prompt plus input image; 1280×720, 5 or 10 seconds at 30 fps. The common paper protocol evaluates half of each task's instances using a fixed random seed.
- Context
- The tables do not specify a score unit or per-model sample count. No percentage interpretation or denominator is inferred.
VGI-Bench visual-intelligence score (Show measurement, test conditions, and source)
- Source value
- 24.5
- Score
- 24.5
- Metric
- Structured Puzzles domain average
- Unit
- No unit provided
- Benchmark version
- arXiv:2608.19583v3 (Table 2)
- Category
- structured-puzzles
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- VGI-Bench, arXiv:2608.19583v3
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Independent evaluation. Image-to-video visual-reasoning tasks; text prompt plus input image; 1280×720, 5 or 10 seconds at 30 fps. The common paper protocol evaluates half of each task's instances using a fixed random seed.
- Context
- The tables do not specify a score unit or per-model sample count. No percentage interpretation or denominator is inferred.
VGI-Bench visual-intelligence score (Show measurement, test conditions, and source)
- Source value
- 47.1
- Score
- 47.1
- Metric
- Physical Manipulation domain average
- Unit
- No unit provided
- Benchmark version
- arXiv:2608.19583v3 (Table 2)
- Category
- physical-manipulation
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- VGI-Bench, arXiv:2608.19583v3
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Independent evaluation. Image-to-video visual-reasoning tasks; text prompt plus input image; 1280×720, 5 or 10 seconds at 30 fps. The common paper protocol evaluates half of each task's instances using a fixed random seed.
- Context
- The tables do not specify a score unit or per-model sample count. No percentage interpretation or denominator is inferred.
H2R-Bench H2RCore (Show measurement, test conditions, and source)
- Source value
- 76.5
- Score
- 76.5
- Metric
- H2RCore score for Parallel-Jaw Gripper
- Unit
- points
- Benchmark version
- arXiv:2608.13049 main evaluation (Tables 2 and S4)
- Category
- human-to-robot-manipulation-transfer
- Direction
- Higher is better
- Lower confidence bound
- 75.7
- Upper confidence bound
- 77.4
- Source type
- independent-evaluator
- Evaluator
- H2R-Bench, arXiv:2608.13049
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Independent evaluation. Human-demonstration to robot-manipulation transfer; Wan2.7 receives the full source video through its native video-conditioned interface; aligned task prompt specifies the parallel-jaw gripper target; generated output 1280×720, 150 frames at 30 fps. Main comparison does not supply a target-robot appearance reference.
- Context
- H2RCore is a task-specific transfer composite, not generic video quality. It weights contact transfer and embodiment correctness at 0.30 each, goal and action completion at 0.15 each, and VBench-based video quality at 0.10. The paper describes a 120-source main set, but no per-row denominator is stored.
H2R-Bench H2RCore (Show measurement, test conditions, and source)
- Source value
- 83.1
- Score
- 83.1
- Metric
- H2RCore score for Dexterous Hand
- Unit
- points
- Benchmark version
- arXiv:2608.13049 main evaluation (Tables 2 and S4)
- Category
- human-to-robot-manipulation-transfer
- Direction
- Higher is better
- Lower confidence bound
- 82.3
- Upper confidence bound
- 83.9
- Source type
- independent-evaluator
- Evaluator
- H2R-Bench, arXiv:2608.13049
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Independent evaluation. Human-demonstration to robot-manipulation transfer; Wan2.7 receives the full source video through its native video-conditioned interface; aligned task prompt specifies the dexterous hand target; generated output 1280×720, 150 frames at 30 fps. Main comparison does not supply a target-robot appearance reference.
- Context
- H2RCore is a task-specific transfer composite, not generic video quality. It weights contact transfer and embodiment correctness at 0.30 each, goal and action completion at 0.15 each, and VBench-based video quality at 0.10. The paper describes a 120-source main set, but no per-row denominator is stored.
Video Models AI-judge benchmark (Show measurement, test conditions, and source)
- Source value
- 3.58
- Score
- 3.58
- Metric
- Weighted overall AI-judge quality score (0-5)
- Unit
- points
- Benchmark version
- Video Models leaderboard updated 2026-10-15
- Category
- video-generation
- Direction
- Higher is better
- Sample
- 64
- Source type
- vendor-reported
- Evaluator
- Video Models by Wiro AI
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Vendor-reported result. Exact Wiro API route alibaba/wan-2-7. Published method states one API run per prompt using model defaults, first successful output, a 5-second duration target, and the resolution closest to 720p. This route documentation lists 5-15s; 720p or 1080p; prompt maximum 5,000 characters. The CSV does not expose the exact per-prompt overrides or criterion-specific denominators.
- Context
- Wiro AI is a commercial model gateway. Scores use its weighted 0-5 AI-judge rubric, not a blind human pairwise preference panel. No per-criterion N or confidence interval is published. This is task-specific AI-judge evidence, not broad independent quality proof.
Recorded measurement details
LMArena Text to Video: LMArena Text to Video (Show measurement, test conditions, and source)
- Source value
- 1337
- Score
- 1337
- Metric
- elo-rating
- Unit
- Elo
- Task
- lmarena-text-to-video
- Task label
- LMArena Text to Video
- Direction
- Higher is better
- Source rank
- 20
- Sample
- 27626
- Lower confidence bound
- 1330
- Upper confidence bound
- 1344
- Retrieved at
- 2026-11-05
LMArena Image to Video: LMArena Image to Video (Show measurement, test conditions, and source)
- Source value
- 1428
- Score
- 1428
- Metric
- elo-rating
- Unit
- Elo
- Task
- lmarena-image-to-video
- Task label
- LMArena Image to Video
- Direction
- Higher is better
- Source rank
- 11
- Sample
- 85521
- Lower confidence bound
- 1423
- Upper confidence bound
- 1433
- Retrieved at
- 2026-11-05
Page 1 of 11
Video sample from our test
This video exists as a documented run for the exact model version. Playback is muted.
Motion and physical consistency
This sample tests Anatomy, object consistency, camera tracking, rain, and reflections.
Prompt and settings
Generate exactly one continuous five-second 16:9 cinematic video. A fictional bicycle courier in a mustard-yellow raincoat rides a red city bicycle from left to right across a rain-wet Hamburg street at blue hour for the entire clip. The camera tracks her smoothly at waist height. Both hands remain visible on the handlebars, the bicycle geometry stays physically correct, and a transparent umbrella fixed to the bicycle moves naturally in the wind. Wet-street reflections and raindrops follow realistic physics. No cuts, no slow motion, no text, no logos, no subtitles, and no audio.
Generated through fal.ai with model fal-ai/wan/v2.7/text-to-video. Archived settings were 16:9, one request, first output, requested 5 s, seed 2037051528, request {"prompt":"Generate exactly one continuous five-second 16:9 cinematic video. A fictional bicycle courier in a mustard-yellow raincoat rides a red city bicycle from left to right across a rain-wet Hamburg street at blue hour for the entire clip. The camera tracks her smoothly at waist height. Both hands remain visible on the handlebars, the bicycle geometry stays physically correct, and a transparent umbrella fixed to the bicycle moves naturally in the wind. Wet-street reflections and raindrops follow realistic physics. No cuts, no slow motion, no text, no logos, no subtitles, and no audio.","aspect_ratio":"16:9","resolution":"720p","duration":5,"enable_prompt_expansion":false}, configured price estimate $0.50.
Popularity compared
How often the model is searched for, used and mentioned, compared with every model of its category.
Popularity index
0 to 100, higher is more popular
20 of 98 models
As of November 2026 · Source: gradually.ai popularity index
Over time
Weekly values from October 18, 2026 to November 5, 2026
Wan 2.7Dreamina Seedance 2.0 720pHunyuanVideo 1.5Veo 3.1 FastKling 2.6 Pro
Wan 2.7 compared
Compare with another model
The most popular comparisons
- 1MiniMax H3 vs. Wan 2.7
- 2Gemini Omni Flash vs. Wan 2.7
- 3FLUX.3 Video vs. Wan 2.7
- 4Dreamina Seedance 2.5 720p vs. Wan 2.7
- 5Dreamina Seedance 2.0 720p vs. Wan 2.7
- 6Sora 2 Pro vs. Wan 2.7
- 7Sora 2 vs. Wan 2.7
- 8Veo 3.1 vs. Wan 2.7
- 9Veo 3.1 Fast vs. Wan 2.7
Show all 48 comparisons
- 10Grok Imagine Video 1.5 vs. Wan 2.7
- 11Grok Imagine Video vs. Wan 2.7
- 12Wan 2.7 vs. Wan 2.6
- 13Wan 2.7 vs. Runway Gen-4.5
- 14Wan 2.7 vs. Kling 3.0 Pro
- 15Wan 2.7 vs. Kling 2.6 Pro
- 16Wan 2.7 vs. Vidu Q3 Pro
- 17Wan 2.7 vs. PixVerse V5.6
- 18Wan 2.7 vs. Hailuo 2.3
- 19Wan 2.7 vs. LTX-2 19B
- 20Wan 2.7 vs. Wan 2.2 A14B
- 21Wan 2.7 vs. HunyuanVideo 1.5
- 22Wan 2.7 vs. Pika 2.2
- 23Wan 2.7 vs. Gemini Omni 1.1 Flash
- 24Wan 2.7 vs. Muse Video
- 25Wan 2.7 vs. Wan 3.0
- 26Wan 2.7 vs. Wan 3.0 Prime
- 27Wan 2.7 vs. HappyHorse 1.0
- 28Wan 2.7 vs. HappyHorse 1.1
- 29Wan 2.7 vs. PixVerse V6
- 30Wan 2.7 vs. PixVerse C1
- 31Wan 2.7 vs. Seedance 2.0 Fast
- 32Wan 2.7 vs. Seedance 2.0 Mini
- 33Wan 2.7 vs. Vidu Q3 Turbo
- 34Wan 2.7 vs. Vidu Q2
- 35Wan 2.7 vs. Vidu Q1
- 36Wan 2.7 vs. Kling 4.0
- 37Wan 2.7 vs. Kling 3.0 Standard API
- 38Wan 2.7 vs. Kling Video O1
- 39Wan 2.7 vs. Hailuo 2.3 Fast
- 40Wan 2.7 vs. Wan 2.5 Preview
- 41Wan 2.7 vs. Wan 2.6 I2V Flash
- 42Wan 2.7 vs. Wan 2.2 TI2V 5B
- 43Wan 2.7 vs. Wan 2.2 I2V A14B
- 44Wan 2.7 vs. HunyuanVideo 1.0 T2V
- 45Wan 2.7 vs. HunyuanVideo I2V 1.0
- 46Wan 2.7 vs. Ray 3.2
- 47Wan 2.7 vs. Grok Imagine Video 1.5 Lite
- 48Wan 2.7 vs. LTX-2.5
All specifications
Input and output
Watermarking and provenance
AI watermarking refers to invisible signals embedded in the output. Visible marks and provenance metadata are listed separately. Unknown means there is insufficient evidence for that particular field.
- AI watermarking
- Unknown
- no source
- Watermarking technique
- Not known
- no source
Licence and access
Prices
No Alibaba subscription includes this model. Through other providers from €6.83 a month (Poe Basic, with limited points). 19 plans in Germany.
| Plan | Monthly price | Note |
|---|---|---|
| Poe Basic | €6.83 | with limited credit |
| Leonardo.Ai Essential | ≈ €12.72 | |
| Magnific Premium | €19.04 | with limited credit |
| Recraft Pro | ≈ €21.20 | with limited credit |
| Recraft Teams | ≈ €23.33 | with limited credit, for teams, per seat |
| Leonardo.Ai Team Starter | ≈ €25.45 | for teams, per seat |
| Poe Plus | €27.36 | with limited credit |
| Leonardo.Ai Premium | ≈ €31.81 | |
| Magnific Premium+ | €42.84 | with limited credit |
| Leonardo.Ai Team Growth | ≈ €50.89 | for teams, per seat |
| Leonardo.Ai Ultimate | ≈ €63.61 | |
| Poe Pro | €68.41 | with limited credit |
| Magnific Business | €70.21 | with limited credit, for teams, per seat |
| Magnific Pro Starter | €104.72 | with limited credit |
| Poe Advanced | €136.84 | with limited credit |
| Magnific Pro | €273.70 | with limited credit |
| Poe Max | €342.11 | with limited credit |
| Leonardo.Ai Custom | n/a | for enterprises |
| Magnific Enterprise | n/a | for enterprises |
Frequently asked questions about Wan 2.7
How good is Wan 2.7?
Rank 14 of 23 in the text-to-video arena rating. With a start image: rank 8 of 21.
LMArena, as of Sep 22, 2026
How much does a video with Wan 2.7 cost?
About $0.75 for a 5 s clip, $0.15 per second of video. That is fairly expensive.
How long can a video with Wan 2.7 be?
Up to 15 seconds in one go. Resolution: 1080p.
According to Alibaba
- Price
- $0.15 per second$0.75 per 5 s clip
- Longest clip
- 15s
- Resolution
- 1080p
- Audio
- yes, generated natively
- Open weights
- no
Where can I use Wan 2.7?
Not in a Alibaba subscription. Including other providers that makes 19 plans in Germany.
Subscription comparison, prices in Germany incl. VAT
At Alibaba
No Alibaba subscription includes this model.
Through other providers
19 plans from 4 providers, from €6.83 a month (Poe Basic), with limited credits.
Sources and data date
Show all 37 sources
Benchmark sources at a glance
Which independent evaluators tested this model. This is coverage, not a quality ranking.
| Evaluator | Test series | Composite indices | Recorded active measurements |
|---|---|---|---|
| Artificial Analysis | 0 | 0 | 3 |
| LMArena | 0 | 0 | 2 |
Without vendor claims and historical or disputed results; those stay labelled in the single values. 5 measurements are not yet assigned to a test series.
| Type | Evidence and data date |
|---|---|
| Research status | Research date October 16, 2026. 29 source URLs checked. This documents the inspected sources, not an exhaustive inventory of every publication. |
| Unresolved | wan-2-7-generic-t2v-score Note The T2V leaderboard's Wan2.7-260612 row is a dated model identifier. No generic Wan 2.7 T2V score is assigned without exact variant attribution. |
| Not found in the inspected sources | arena-video-edit-exact-row Note The inspected Video Edit table has no exact model-version row. T2V and I2V values were not reused. |
| Not found in the inspected sources | vabench-exact-model-row Note VABench has no exact model row. Sora 2 was not transferred to Sora 2 Pro or other models. |
| Not found in the inspected sources | artificial-analysis-t2v-exact-row Note No exact model version or resolution was found in the inspected Artificial Analysis table. |
| Unresolved | vbench-current-public-leaderboard Note The current VBench leaderboard is embedded in a dynamic Hugging Face iframe and was not retrievable; current scores remain unresolved. |
| Unresolved | vidbenchmark-performance-axes Note The source publishes speed-to-first-frame and cost-efficiency axes, but its inspected methodology does not define them as measured performance results or disclose units, direction, or a scoring formula. They are omitted. Model limits, maximum resolution, and list prices are not quality measurements. |
| Not published | AA and VVA output resolution Note The AA image-to-video row and VVA-Bench safety table do not publish model output resolution; CaliBench's Wan2.7 row is explicitly 720p. |
| Not published | servingRevision Original wording The dataset identifies the participant name but not provider API route, served checkpoint revision, quantization, or resolution. |
| Unresolved | datasetRowCount Original wording The pinned README states 21,125 matchup rows, while current Hugging Face viewer metadata reports 22,625. Neither aggregate is used as sampleCount or leaderboard N. |
| Additional source | Alibaba · Alibaba Cloud Model Studio video generation and editing · Editorial description reviewed October 2, 2026 |
| Additional source | Video pricing |
| Additional source | fal.ai |
| Additional source | LMArena (retrieved November 5, 2026)12 |
| Additional source | Artificial Analysis AA-Video-I2V v1.1 (retrieved October 16, 2026; October 17, 2026) |
| Additional source | VGI-Bench, arXiv:2608.19583v3 (retrieved October 16, 2026) |
| Additional source | H2R-Bench, arXiv:2608.13049 (retrieved October 16, 2026) |
| Additional source | Video Models by Wiro AI (retrieved October 16, 2026; October 28, 2026) |
| Additional source | vidbenchmark (retrieved October 16, 2026) |
| Additional source | CaliBench, Table 2 (retrieved October 17, 2026; October 18, 2026) |
| Additional source | VVA-Bench, VPA-Guard paper Table 5 (retrieved October 17, 2026) |
| Additional source | Rapidata, image-to-video-physics-human-feedback-benchmark dataset README (retrieved October 17, 2026) |
| Additional source | Ego2Act: Evaluating Goal-Directed Manipulation in Egocentric Video Generation (retrieved October 17, 2026) |
| Additional source | ReCA project page, paper Table 1 (retrieved October 18, 2026) |
| Additional source | WBench paper, arXiv:2605.25874 (retrieved October 22, 2026) |
| Additional source | VGIF-Score paper, independent evaluator (retrieved October 22, 2026) |
| Additional source | WBench official leaderboard (retrieved October 23, 2026) |
| Additional source | Arena Video Image-to-Video (retrieved September 21, 2026; October 16, 2026; October 18, 2026; October 28, 2026) |
| Additional source | Arena Video Text-to-Video (retrieved September 21, 2026; October 16, 2026; October 18, 2026; October 28, 2026) |
| Additional source | Artificial Analysis AA-Video-Editing v1.1 (retrieved October 28, 2026) |
| Additional source | MemeBench Video Arena (retrieved October 28, 2026) |
| Additional source | Rapidata I2V Physics benchmark (retrieved October 22, 2026; October 28, 2026) |
| Additional source | Model metadata (retrieved October 16, 2026)123 |
| Additional source | Model metadata (retrieved October 16, 2026; October 28, 2026) |
| Additional source | Model metadata (retrieved October 17, 2026) |
| Additional source | Model metadata (retrieved October 18, 2026)12 |
| Additional source | Model metadata (retrieved October 28, 2026) |
| Additional source | OpenRouter: Popularity index (derived), CC BY 4.0 |
| Additional source | Vercel AI Gateway: Popularity index (derived), CC BY 4.0 |
| Additional source | Ollama: cumulative downloads for the popularity index |
| Additional source | Replicate: cumulative runs for the popularity index |
| Additional source | Google Trending Now: limited feed selection for Germany and the US |
| Additional source | Creative Commons Attribution 4.0 |