Seedance 2.5: Video Leaderboard Podium and Real API Costs

👁 113 reads

Seedance 2.5 from ByteDance holds 3rd place on the Artificial Analysis text-to-video leaderboard with 1,143 Elo points - just 13 behind the leader, Wan 3.0. Yet that same table lists an eye-opening figure: $34.12 per minute of video. That makes it the most expensive model in the top ten, nearly triple the leader and seven times the cost of MiniMax H3. I looked into why that discrepancy exists and what Seedance 2.5 actually costs to run. Full guide: Seedance.

Top 10: Quality and Pricing

Source: Artificial Analysis AA-Video-T2V v2.0, October 11, 2026. Seedance 2.5 gathered 8,264 blind comparisons, and its confidence interval (1134-1152) overlaps with Utopai X and H3, effectively putting places 2 through 4 in a dead heat.

What Seedance 2.5 Actually Costs

Seedance does not charge per second. You pay for video tokens: width x height x frame count / 1024. I verified this formula directly against my own usage invoices - my 5-second Seedance 2.5 clip in 480p consumed 48,440 tokens and cost $0.5183 on OpenRouter, precisely matching the math. Using the same formula:

• One minute at 480p on OpenRouter ($10.70 per million tokens): roughly $6.20;
• One minute at 720p: roughly $13.90;
• On fal.ai, the token rate is doubled ($21.40 per million) - my 5-second 480p clip cost $1.06 there;
• The $34.12 figure from the AA leaderboard therefore reflects a higher-tier setup: higher resolution or a more expensive host. AA does not specify which configuration it sampled.

The practical takeaway: the price you pay for Seedance 2.5 depends far more on the resolution and the provider you choose than on the model itself. The difference between the cheapest and priciest routes spans several multiples.

What You Get for the Price

ByteDance rolled out Seedance 2.5 on July 31, 2026 across Dreamina and Chinese apps Jimeng and Doubao, followed by the BytePlus API on August 6. Key capabilities include:

• 30 seconds in a single pass, across multiple shots, with multi-extension support - scalable to several minutes according to ByteDance;
• Up to 50 concurrent reference inputs: 30 images, 10 video clips, 10 audio tracks, plus 3D mockups for camera staging;
• Timestamp-based regional editing, green-screen replacement, and camera motion adjustments on completed clips.

No other leading model currently combines all three workflows. Official demo clips with prompts - a concert with 18 references, a Peking opera scene, and an animation built from a 3D blockout - are detailed in What Seedance 2.5 Can Do.

Where Seedance 2.5 Falls Short

In my Video Arenaacross five short 480p test scenes, Seedance 2.5 averaged a score of 5.6/10, compared to 8.9 for FLUX 3 and 8.8 for Wan 3.0. The steepest drops came in non-English evaluation: we tested it in Polish, where rendering the sign (Polish original translated: “YELLOW BOAT”) produced corrupted lettering in the original Polish test, and it misidentified more than half the words in a Polish spoken sentence (WER 0.583). Polish is notably absent from the model’s official list of 11 supported languages. Because the AA leaderboard benchmarks general visual quality using English prompts, this limitation never surfaces there.

Key Takeaways

• For complex scenes with heavy referencing, Seedance 2.5 has no direct rival, justifying its premium.
• For quick cuts and non-English text, Wan 3.0 or FLUX 3 Video are better suited - scoring higher in my testing while costing less.
• To run Seedance 2.5 at minimal cost, stick to 480p via OpenRouter or check regional plans on Dreamina.

Top-tier comparison: Seedance 2.5 vs Wan 3.0 vs Utopai X. Full model cost index: how much AI video costs per second.

// read moreSeedance 2.5 in Action: 8 Prompts and Video Examples TestedAI ToolsSeedance 2.5 vs Wan 3.0 vs Utopai X: prices and video resultsMagazineSeedance 2.5: Features, Pricing & Unlimited Access GuideAI Tools