OpenArt tops the anime video ranking, and its rival costs eight times more
Physion-Arc 1.0 Anime Track scored eight agents on 19 metrics. OpenArt leads with 77.8. Read the cost column and the table shifts: Invideo charges $50.7 per video for 69.2 points, while MiniMax delivers 68.5 for $6.3.
Physion Labs has added an Anime Track to Physion-Arc 1.0, its benchmark for video agents. Eight agents were run on the same 100 anime prompts and every video was scored on 19 metrics. The agents were tested across three input modes: text only, single reference image, and multiple reference images.
The ranking
| # | Agent | Overall quality | Cost per video |
|---|---|---|---|
| 1 | OpenArt | 77.8 | $21.3 |
| 2 | Anime Agent 1.0 | 75.6 | $25.0 |
| 3 | Invideo Agent One | 69.2 | $50.7 |
| 4 | Runway Agent 2.0 | 68.7 | $10.5 |
| 5 | MiniMax Hub v2.0.2 | 68.5 | $6.3 |
| 6 | Luma Creative Agents | 64.3 | $16.4 |
| 7 | Flova 1.0 | 63.8 | $2.5 |
| 8 | Utopai PAI 2.0 | 61.3 | $20.4 |
Read the cost column
Read the ranking alone and third place looks good. Read it with the price and something else appears: Invideo Agent One charges $50.7 per video for 69.2 points, while MiniMax Hub returns 68.5 for $6.3. Eight times the money for 0.7 of a point. In the same vein, Flova 1.0 scores 63.8 at $2.5, so for anyone willing to give up 14 points to save 19 dollars a video, the table inverts.
Three dimensions, not one winner
The metrics group into three headings: narrative coherence (7 metrics), animetic language (7 metrics) and production quality (5 metrics). OpenArt takes the overall lead, but narrative coherence is led by Anime Agent 1.0, 82.9 against 82.3. OpenArt is clearly ahead on animetic language (68.6) and production quality (84.5).
The distinction matters. Narrative coherence measures whether a character stays the same character and whether a prop survives from scene to scene. Animetic language measures directorial choices: camera, composition, pacing, sound. An agent can tell a coherent story and direct it badly, or the reverse.
Half the metrics are subjective
The board labels every metric objective or subjective, which is an honest split. The objective side is high and tightly bunched: on technical coherence all eight sit between 91.8 and 99.3, and on audio stability everyone except the last-placed 48.1 is above 77. The spread opens on the subjective metrics: emotional intent runs from 38.6 to 62.9, audio integration from 31.6 to 67.9. Technically the agents have converged; where they separate is direction.
Physion Labs says that when a new agent joins the pool it recomputes results across all 100 prompts, because an expanded comparison set shifts the subjective scores. Worth keeping in mind while reading: these numbers are not absolute, they are relative to this pool of eight.
