ByteDance opens its cheaper Seedance 2.5 Flash video model to API users

ByteDance opened API access to Dreamina Seedance 2.5 Flash on BytePlus ModelArk on November 15, pricing the faster video model at $7.0 per million tokens for 480p and 720p output versus $10.70 for the full Seedance 2.5, according to CellCog's breakdown of the BytePlus announcement. A five-second 720p clip costs about $0.76, or roughly $0.15 per second, and PANews reported Volcano Engine prices 720p text-to-video and image-to-video at about 0.99 yuan per second, around 34% below Seedance 2.5, with generation speed roughly doubled. Flash keeps the 30-second clip length and built-in audio but drops 1080p, 4K and offline inference, and caps inputs at 10 images, 5 videos and 5 audio clips (20 files) against the full model's 50.

A five-second AI video clip now costs 76 cents on ByteDance's new fast tier, and the catch is that you can't get 1080p. If you build video tools on someone else's model, the trade matters more than the discount. Dreamina Seedance 2.5 Flash is the version of ByteDance's video model that most startups will actually be able to afford to call in bulk. According to CellCog's breakdown of the BytePlus announcement, the model costs $7.0 per million tokens at 480p and 720p, against $10.70 for the full Seedance 2.5, and a five-second 720p clip works out to about $0.76, or roughly $0.15 per second. ByteDance announced Flash on November 12, and the API opened on BytePlus ModelArk today, November 15. Chinese-language coverage tells the same story in yuan. PANews reported that Volcano Engine, ByteDance's cloud arm, prices text-to-video and image-to-video at 720p at about 0.99 yuan per second, around 34% below Seedance 2.5, with generation speed roughly doubled. Consumer access started on November 12 through Dreamina, LibTV and Updream, per the same report. The full Seedance 2.5 launched on July 31 with native 1080p, clips of up to 30 seconds with built-in audio, and room for up to 50 reference files in one generation, as The Decoder and BytePlus's own documentation describe it. Flash keeps the 30-second length and the audio. It drops the top of the resolution ladder: the ModelArk documentation, as summarised by CellCog, lists 1080p, 4K and offline inference as unsupported, so you get 480p and 720p only. Reference limits shrink too. PANews says a Flash input can carry up to 10 images, 5 videos and 5 audio clips, which is 20 files, well under the full model's 50. ByteDance still says Flash edits video and beats the previous flagship, Seedance 2.0, on instruction following, motion and audio, and that its output is the closest of any ByteDance model to Seedance 2.5. Fal Doubles To A $16 Billion Valuation As AI Inference Demand Soars https://startupfortune.com/fal-doubles-to-a-16-billion-valuation-as-ai-inference-demand-soars/ Bessemer Venture Partners led the $1.3 billion round, with Nvidia, Sound Ventures and T. Rowe Price also investing in a startup whose CEO admits it had to rate-limit customers this summer. - AI inference company valuation doubles to $16 billion https://startupfortune.com/fal-doubles-to-a-16-billion-valuation-as-ai-inference-demand-soars/ - fastest growing AI infrastructure startup funding round https://startupfortune.com/fal-doubles-to-a-16-billion-valuation-as-ai-inference-demand-soars/ Treat the speed claim with some caution. BytePlus says Flash runs at "nearly twice the generation speed" of the flagship, but CellCog noted the documentation gave no timing measurements, and it hadn't tested the model itself when it published. Quality is the same story: ByteDance's wording is "closest," not "equal." Who should use which Frankly, the split is easy. Draft with Flash, finish with the full model. If your product lets a shopper or a creator try six takes of a 15-second product clip before picking one, the 34% saving and the faster turnaround land on the exact step where you burn the most money, and 720p is fine for a preview. If the client is going to run the final cut on a broadcast screen or a paid social placement that wants 1080p, only the full model can deliver it, and the extra reference files help when a brand needs the same character or logo to hold steady across shots. That workflow also changes what you can promise customers. A per-clip price near 76 cents makes a free-trial tier or a flat-rate plan easier to model than it was at the flagship rate, though you'd want to check real usage before you pick a price. ByteDance hasn't said whether Flash gets any volume discount, and CellCog's write-up flagged that geographic availability and benchmarks were not stated either. The pressure on OpenAI, Google and Kling The rivals haven't answered at this tier yet. No like-for-like cheaper, faster 30-second model with native audio has been announced by OpenAI, Google or Kuaishou's Kling, so ByteDance is setting the reference price for this class of clip. That's a problem for any Western lab selling video at a premium, because the startups buying from them can read a price sheet as well as you can. Many of them already route calls through aggregators, which makes swapping a model a configuration change rather than a rewrite. One caveat on all of this. The price figures come from ByteDance's own platforms and secondary write-ups of them, not from independent testing, and the speed gain remains a vendor claim. Run your own prompts through both models before you move a production workload. Also read: How to Calculate AI Agent Headcount Offset Before You Post the Job https://startupfortune.com/how-to-calculate-ai-agent-headcount-offset-before-you-post-the-job/ • Anthropic's $64 billion IPO lands in the middle of an AI selloff https://startupfortune.com/anthropics-64-billion-ipo-lands-in-the-middle-of-an-ai-selloff/ • Congress moves to tighten AI chip export rules after Jensen Huang quits Nvidia https://startupfortune.com/congress-moves-to-tighten-ai-chip-export-rules-after-jensen-huang-quits-nvidia/ This article is posted in AI News https://startupfortune.com/category/ai/ , check it out for more related stories. Join the discussion Open in the community → https://startupfortune.com/community/ Almost there. Sign in and your reply posts straight away.