Tencent HunyuanVideo modelOpen weights
Hy-Nomas
Summary
Hy-Nomas generates an interactive world through successive video segments. Text instructions, camera paths, and prior history guide which objects, actions, and viewpoints appear in the next segment. Its diffusion backbone has eight billion parameters. The released standard configuration renders 720 × 1280 pixels at 24 frames per second and uses eight DiT denoising steps.
Tencent evaluates interaction and spatial consistency in 280 test cases using pairwise human judgments. The results are win, tie, and loss rates against named models, not universal quality scores. The LoRA ablation evaluates a fine-tuned cinematic variant. Hy-Nomas-PE adds prompt enhancement to the base model; results for each refer to that variant. The 24 frames per second value describes video playback rate, not how quickly the model generates it.
All benchmark results
Every published result with its source, test setup and comparison models.
Show all 78 published benchmark results
Page 1 of 7
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.
Bench-1 multi-turn text interaction prompt following vs HappyOyster-Directing: win (Show measurement, test conditions, and source)
- Source value
- 65
- Score
- 65
- Metric
- pairwise human evaluation win rate
- Unit
- %
- Benchmark version
- Hy-Nomas-Bench Bench-1 multi-turn text interaction
- Category
- world-model-human-pairwise
- Direction
- Higher is better
- Sample
- 60
- Source type
- vendor-reported
- Evaluator
- Tencent Hy-Nomas technical report, human evaluation figures
- Status
- active
- Retrieved at
- 2026-10-29
- Methodology
- Vendor-reported result. Bench-1 multi-turn text interaction; Hy-Nomas vs HappyOyster-Directing; prompt following; perspective of Hy-Nomas; three trained annotators per evaluation group; 60 benchmark cases. Benchmark cases are evaluation examples, not model runs.
- Context
- Win, tie, and loss are separate shares of pairwise human judgments, not a standalone benchmark score. The case count is the number of benchmark examples, not model runs.
Bench-1 multi-turn text interaction prompt following vs HappyOyster-Directing: tie (Show measurement, test conditions, and source)
- Source value
- 25
- Score
- 25
- Metric
- pairwise human evaluation tie rate
- Unit
- %
- Benchmark version
- Hy-Nomas-Bench Bench-1 multi-turn text interaction
- Category
- world-model-human-pairwise
- Direction
- No direction published
- Sample
- 60
- Source type
- vendor-reported
- Evaluator
- Tencent Hy-Nomas technical report, human evaluation figures
- Status
- active
- Retrieved at
- 2026-10-29
- Methodology
- Vendor-reported result. Bench-1 multi-turn text interaction; Hy-Nomas vs HappyOyster-Directing; prompt following; perspective of Hy-Nomas; three trained annotators per evaluation group; 60 benchmark cases. Benchmark cases are evaluation examples, not model runs.
- Context
- Win, tie, and loss are separate shares of pairwise human judgments, not a standalone benchmark score. The case count is the number of benchmark examples, not model runs.
Bench-1 multi-turn text interaction prompt following vs HappyOyster-Directing: loss (Show measurement, test conditions, and source)
- Source value
- 10
- Score
- 10
- Metric
- pairwise human evaluation loss rate
- Unit
- %
- Benchmark version
- Hy-Nomas-Bench Bench-1 multi-turn text interaction
- Category
- world-model-human-pairwise
- Direction
- Lower is better
- Sample
- 60
- Source type
- vendor-reported
- Evaluator
- Tencent Hy-Nomas technical report, human evaluation figures
- Status
- active
- Retrieved at
- 2026-10-29
- Methodology
- Vendor-reported result. Bench-1 multi-turn text interaction; Hy-Nomas vs HappyOyster-Directing; prompt following; perspective of Hy-Nomas; three trained annotators per evaluation group; 60 benchmark cases. Benchmark cases are evaluation examples, not model runs.
- Context
- Win, tie, and loss are separate shares of pairwise human judgments, not a standalone benchmark score. The case count is the number of benchmark examples, not model runs.
Bench-1 multi-turn text interaction physical plausibility vs HappyOyster-Directing: win (Show measurement, test conditions, and source)
- Source value
- 45
- Score
- 45
- Metric
- pairwise human evaluation win rate
- Unit
- %
- Benchmark version
- Hy-Nomas-Bench Bench-1 multi-turn text interaction
- Category
- world-model-human-pairwise
- Direction
- Higher is better
- Sample
- 60
- Source type
- vendor-reported
- Evaluator
- Tencent Hy-Nomas technical report, human evaluation figures
- Status
- active
- Retrieved at
- 2026-10-29
- Methodology
- Vendor-reported result. Bench-1 multi-turn text interaction; Hy-Nomas vs HappyOyster-Directing; physical plausibility; perspective of Hy-Nomas; three trained annotators per evaluation group; 60 benchmark cases. Benchmark cases are evaluation examples, not model runs.
- Context
- Win, tie, and loss are separate shares of pairwise human judgments, not a standalone benchmark score. The case count is the number of benchmark examples, not model runs.
Bench-1 multi-turn text interaction physical plausibility vs HappyOyster-Directing: tie (Show measurement, test conditions, and source)
- Source value
- 43.3
- Score
- 43.3
- Metric
- pairwise human evaluation tie rate
- Unit
- %
- Benchmark version
- Hy-Nomas-Bench Bench-1 multi-turn text interaction
- Category
- world-model-human-pairwise
- Direction
- No direction published
- Sample
- 60
- Source type
- vendor-reported
- Evaluator
- Tencent Hy-Nomas technical report, human evaluation figures
- Status
- active
- Retrieved at
- 2026-10-29
- Methodology
- Vendor-reported result. Bench-1 multi-turn text interaction; Hy-Nomas vs HappyOyster-Directing; physical plausibility; perspective of Hy-Nomas; three trained annotators per evaluation group; 60 benchmark cases. Benchmark cases are evaluation examples, not model runs.
- Context
- Win, tie, and loss are separate shares of pairwise human judgments, not a standalone benchmark score. The case count is the number of benchmark examples, not model runs.
Bench-1 multi-turn text interaction physical plausibility vs HappyOyster-Directing: loss (Show measurement, test conditions, and source)
- Source value
- 11.7
- Score
- 11.7
- Metric
- pairwise human evaluation loss rate
- Unit
- %
- Benchmark version
- Hy-Nomas-Bench Bench-1 multi-turn text interaction
- Category
- world-model-human-pairwise
- Direction
- Lower is better
- Sample
- 60
- Source type
- vendor-reported
- Evaluator
- Tencent Hy-Nomas technical report, human evaluation figures
- Status
- active
- Retrieved at
- 2026-10-29
- Methodology
- Vendor-reported result. Bench-1 multi-turn text interaction; Hy-Nomas vs HappyOyster-Directing; physical plausibility; perspective of Hy-Nomas; three trained annotators per evaluation group; 60 benchmark cases. Benchmark cases are evaluation examples, not model runs.
- Context
- Win, tie, and loss are separate shares of pairwise human judgments, not a standalone benchmark score. The case count is the number of benchmark examples, not model runs.
Bench-1 multi-turn text interaction prompt following vs LingBot-World v2: win (Show measurement, test conditions, and source)
- Source value
- 75
- Score
- 75
- Metric
- pairwise human evaluation win rate
- Unit
- %
- Benchmark version
- Hy-Nomas-Bench Bench-1 multi-turn text interaction
- Category
- world-model-human-pairwise
- Direction
- Higher is better
- Sample
- 60
- Source type
- vendor-reported
- Evaluator
- Tencent Hy-Nomas technical report, human evaluation figures
- Status
- active
- Retrieved at
- 2026-10-29
- Methodology
- Vendor-reported result. Bench-1 multi-turn text interaction; Hy-Nomas vs LingBot-World v2; prompt following; perspective of Hy-Nomas; three trained annotators per evaluation group; 60 benchmark cases. Benchmark cases are evaluation examples, not model runs.
- Context
- Win, tie, and loss are separate shares of pairwise human judgments, not a standalone benchmark score. The case count is the number of benchmark examples, not model runs.
Bench-1 multi-turn text interaction prompt following vs LingBot-World v2: tie (Show measurement, test conditions, and source)
- Source value
- 16.7
- Score
- 16.7
- Metric
- pairwise human evaluation tie rate
- Unit
- %
- Benchmark version
- Hy-Nomas-Bench Bench-1 multi-turn text interaction
- Category
- world-model-human-pairwise
- Direction
- No direction published
- Sample
- 60
- Source type
- vendor-reported
- Evaluator
- Tencent Hy-Nomas technical report, human evaluation figures
- Status
- active
- Retrieved at
- 2026-10-29
- Methodology
- Vendor-reported result. Bench-1 multi-turn text interaction; Hy-Nomas vs LingBot-World v2; prompt following; perspective of Hy-Nomas; three trained annotators per evaluation group; 60 benchmark cases. Benchmark cases are evaluation examples, not model runs.
- Context
- Win, tie, and loss are separate shares of pairwise human judgments, not a standalone benchmark score. The case count is the number of benchmark examples, not model runs.
Bench-1 multi-turn text interaction prompt following vs LingBot-World v2: loss (Show measurement, test conditions, and source)
- Source value
- 8.3
- Score
- 8.3
- Metric
- pairwise human evaluation loss rate
- Unit
- %
- Benchmark version
- Hy-Nomas-Bench Bench-1 multi-turn text interaction
- Category
- world-model-human-pairwise
- Direction
- Lower is better
- Sample
- 60
- Source type
- vendor-reported
- Evaluator
- Tencent Hy-Nomas technical report, human evaluation figures
- Status
- active
- Retrieved at
- 2026-10-29
- Methodology
- Vendor-reported result. Bench-1 multi-turn text interaction; Hy-Nomas vs LingBot-World v2; prompt following; perspective of Hy-Nomas; three trained annotators per evaluation group; 60 benchmark cases. Benchmark cases are evaluation examples, not model runs.
- Context
- Win, tie, and loss are separate shares of pairwise human judgments, not a standalone benchmark score. The case count is the number of benchmark examples, not model runs.
Bench-1 multi-turn text interaction physical plausibility vs LingBot-World v2: win (Show measurement, test conditions, and source)
- Source value
- 68.3
- Score
- 68.3
- Metric
- pairwise human evaluation win rate
- Unit
- %
- Benchmark version
- Hy-Nomas-Bench Bench-1 multi-turn text interaction
- Category
- world-model-human-pairwise
- Direction
- Higher is better
- Sample
- 60
- Source type
- vendor-reported
- Evaluator
- Tencent Hy-Nomas technical report, human evaluation figures
- Status
- active
- Retrieved at
- 2026-10-29
- Methodology
- Vendor-reported result. Bench-1 multi-turn text interaction; Hy-Nomas vs LingBot-World v2; physical plausibility; perspective of Hy-Nomas; three trained annotators per evaluation group; 60 benchmark cases. Benchmark cases are evaluation examples, not model runs.
- Context
- Win, tie, and loss are separate shares of pairwise human judgments, not a standalone benchmark score. The case count is the number of benchmark examples, not model runs.
Bench-1 multi-turn text interaction physical plausibility vs LingBot-World v2: tie (Show measurement, test conditions, and source)
- Source value
- 26.7
- Score
- 26.7
- Metric
- pairwise human evaluation tie rate
- Unit
- %
- Benchmark version
- Hy-Nomas-Bench Bench-1 multi-turn text interaction
- Category
- world-model-human-pairwise
- Direction
- No direction published
- Sample
- 60
- Source type
- vendor-reported
- Evaluator
- Tencent Hy-Nomas technical report, human evaluation figures
- Status
- active
- Retrieved at
- 2026-10-29
- Methodology
- Vendor-reported result. Bench-1 multi-turn text interaction; Hy-Nomas vs LingBot-World v2; physical plausibility; perspective of Hy-Nomas; three trained annotators per evaluation group; 60 benchmark cases. Benchmark cases are evaluation examples, not model runs.
- Context
- Win, tie, and loss are separate shares of pairwise human judgments, not a standalone benchmark score. The case count is the number of benchmark examples, not model runs.
Bench-1 multi-turn text interaction physical plausibility vs LingBot-World v2: loss (Show measurement, test conditions, and source)
- Source value
- 5
- Score
- 5
- Metric
- pairwise human evaluation loss rate
- Unit
- %
- Benchmark version
- Hy-Nomas-Bench Bench-1 multi-turn text interaction
- Category
- world-model-human-pairwise
- Direction
- Lower is better
- Sample
- 60
- Source type
- vendor-reported
- Evaluator
- Tencent Hy-Nomas technical report, human evaluation figures
- Status
- active
- Retrieved at
- 2026-10-29
- Methodology
- Vendor-reported result. Bench-1 multi-turn text interaction; Hy-Nomas vs LingBot-World v2; physical plausibility; perspective of Hy-Nomas; three trained annotators per evaluation group; 60 benchmark cases. Benchmark cases are evaluation examples, not model runs.
- Context
- Win, tie, and loss are separate shares of pairwise human judgments, not a standalone benchmark score. The case count is the number of benchmark examples, not model runs.
Page 1 of 7
All specifications
Architecture
Licence and access
Further details
- Interaction
- Multi-turn free-form text instructions and camera control; both can be combined in one session
- Source: Interaction
- Standard output
- 720 × 1280 pixels at 24 frames per second and eight DiT denoising steps. The 24 FPS value is playback frame rate, not generation speed.
- Source: Standard output
- Single-GPU configuration
- One NVIDIA GeForce RTX 4090; INT8 DiT and KV storage, FP8 PixelUMM decoder, CPU offload, and bounded GPU history
- Source: Single-GPU configuration
- Hy-Nomas-Bench
- 280 test cases across four tasks: multi-turn text interaction, camera navigation, interleaved text and camera control, and spatial memory. Cases include interactions requiring videos longer than 60 seconds.
- Source: Hy-Nomas-Bench
- Evaluation method
- Pairwise human evaluation reported as win, tie, and loss rates; three trained annotators per evaluation group. Benchmark cases are not model runs.
- Source: Evaluation method
- Standard video output
- 720 × 1280 pixels, 24 FPS, eight DiT denoising steps (24 FPS is the rendered video frame rate and does not measure generation speed.)
- Source: Standard video output
- Low-memory inference
- Single NVIDIA GeForce RTX 4090; INT8 DiT and KV storage; FP8 decoder; CPU offload
- Source: Low-memory inference
- Benchmark case design
- 280 test cases: 60 multi-turn text, 100 camera navigation, 60 interleaved control, 60 spatial memory
- Source: Benchmark case design
Sources and data date
Benchmark sources at a glance
Which independent evaluators tested this model. This is coverage, not a quality ranking.
No active result from the reviewed external evaluators can currently be attributed to this model. Other source evidence remains available below.
Without vendor claims and historical or disputed results; those stay labelled in the single values.
| Type | Evidence and data date |
|---|---|
| Research status | Research date October 29, 2026. 60 source URLs checked. This documents the inspected sources, not an exhaustive inventory of every publication. |
| Not published | Robotics LoRA numeric benchmark Note The report describes robotics-adapter examples qualitatively but gives no numerical evaluation table for the robotics LoRA. |
| Not published | Cinematic visual quality score Note The cinematic adapter is discussed qualitatively; the numeric ablation covers prompt following, physical plausibility, and camera control. |
| Not published | Hy-Nomas-PE evaluation case count Note Figure 21 reports rank-gap-weighted pairwise human rates but does not state the number of evaluated cases. |
| Not published | Exact generation throughput Note 24 FPS is the video's output frame rate, not generation speed. The report does not provide an exact general generation-throughput benchmark. |
| Not found in the inspected sources | Vals rows in captured payloads Note Nine complete Vals payloads dated 28 October were checked; no exact model ID or unambiguous model label. Limited to these saved benchmarks. |
| Unresolved | Uncaptured Arena Code/Speech leaderboards Note Text, Agent including Code/Chat/Work, and twelve WebDev views were captured. A separate general Code and Arena Speech leaderboard were not saved as complete tables. |
| Unresolved | Artificial Analysis catalog reconciliation Note The full current AA model and benchmark catalog was not re-reconciled for this profile in this sidecar. This sidecar makes no AA absence claim; native AA imports remain authoritative. |
| Additional source | Tencent Hy-Nomas model card (retrieved October 29, 2026) · Tencent Hy-Nomas model card · Editorial description reviewed October 29, 2026 |
| Additional source | Tencent Hy-Nomas release repository (retrieved October 29, 2026) · Tencent Hy-Nomas release repository · Editorial description reviewed October 29, 2026 |
| Additional source | Hy-Nomas technical report (retrieved October 29, 2026) · Hy-Nomas technical report · Editorial description reviewed October 29, 2026 |
| Additional source | Hy-Nomas-Bench dataset (retrieved October 29, 2026) · Hy-Nomas-Bench dataset · Editorial description reviewed October 29, 2026 |
| Additional source | Model metadata (retrieved October 29, 2026)12 |