Topview shipped its first in-house video model on 21 October 2026, with a headline aimed directly at the model it has been reselling: "Built to rival Seedance 2.5 1080P quality at 62% lower cost."
That is a big number, and it is roughly right on the vendor's own arithmetic — T1 is listed at 38% of Seedance's official price. But if you are the person wiring a video model into a product, the number that matters is not 62%. It is the number you get when you plug your shot mix into it. For most pipelines that number is meaningfully smaller, and for one common workload it is negative.
Here is what I could verify, and the gotchas that only show up after you have committed credits.
TL;DR
- T1 is priced at roughly $0.22/s at 1080p against roughly $0.585/s for Seedance 2.5, per third-party records. Topview bills in credits, so recompute from your own invoice before you plan anything.
- An independent 10-challenge head-to-head puts T1 ahead on lip-sync, character consistency and cinematic color, behind on complex physics, fight choreography, fast action, on-screen text and prompt adherence. Dance and choreography still go to Seedance 2.5.
- The 62% is a ceiling. A 65/35 split across shot types lands closer to 41%.
- The "30 days unlimited" tier is 720p, standard queue, web-only, and explicitly excluded from API/MCP. If you ship on the API, it does not touch your cost of goods.
What actually shipped
Topview is a Singapore-based company behind an agent-style video workspace — Canvas, Drama Studio, a 3D shot composer, character swap — that routes requests across several third-party models. T1 is its first proprietary model. No architecture details have been published.
The detail that reframes the announcement: Topview already resells Seedance 2.5. T1 is not just another model in the picker; it is a vendor shipping a cheaper substitute for something it currently pays another company for. The 62% is a margin structure being rewritten, which is why it is worth reading carefully rather than adopting on day one. The launch post itself pulled 280K views and ranked 1,362 of 2,071 AI launches in the directory that tracks these — enough to be argued about, not enough to call it a default.
| Reported 1080p rate | vs Seedance 2.5 | |
|---|---|---|
| Topview T1 | ~$0.22 / second | −62% |
| Seedance 2.5 | ~$0.585 / second | — |
Two caveats on that table, both material. The per-second figure comes from third-party review records, one of which flags its own pricing as un-re-verified since 23 October — Topview does not publish a clean $/second. And because billing runs through credits that are discounted differently per plan and resolution, the same clip can cost very different amounts depending on whether you are on Plus, Business or Ultra annual. Treat every $/second number, including mine, as an estimate to be replaced by your invoice.
For reference, the plans as listed: Plus $18/month, Business $44/month, Ultra $50/month, Team $56 per seat, all on annual billing. No free tier is documented for T1.
Where it wins and where it loses
One independent comparison ran T1 and Seedance 2.5 across ten challenges — lip-sync, dialogue, physics, fight scenes, dancing, prompt adherence among them — self-funded despite a past sponsorship relationship with the company. The result is not a wash, and it is not a T1 victory lap:
| Capability | Better model |
|---|---|
| Lip-sync, including non-English | T1 |
| Character consistency across a scene | T1 |
| Cinematic color, camera movement | T1 (crisper, more saturated) |
| Complex physics (e.g. glass shattering) | Seedance 2.5 |
| Fight choreography | Seedance 2.5 |
| Fast action (T1 blurs) | Seedance 2.5 |
| On-screen text / signage | Seedance 2.5 (T1 renders it illegible) |
| Prompt adherence | Seedance 2.5 |
| Dance and choreography | Seedance 2.5 |
The pattern is not noise. T1 is strongest when the camera is calm and a face is doing something human; it is weakest when a body has to do something physically unlikely and land it. That split is the whole decision. If your workload is talking-head ads, product b-roll, dialogue scenes, avatar explainers — the discount is real and you keep almost all the quality. If your workload is dance clips, trend remakes, fight beats, or anything where a contact point has to be believable, the cheaper model is a downgrade with a smaller invoice, and you will spend the difference on retries.
Three things that bite after you commit
1. Unlimited is 720p, and the headline is 1080p. The pricing page line item reads Topview T1, 720P, 30 Days Unlimited on Ultra annual. Nothing wrong with that — but if you benchmarked the 1080p comparison and then budgeted on the unlimited tier, you are planning two different products.
2. Unlimited does not reach your pipeline. Topview's terms state that unlimited and free generations on plans are available only on the website and are not accessible via API, MCP, CLI or other automation. Unlimited also means credit-free, not priority — it runs in the standard queue, and wait times vary with demand and recent usage. Fine for a research window. Not something to put an SLA on.
3. The credit wallets are not interchangeable. Plus and Business plan credits can pay for API calls. Ultra and Team credits cannot. Monthly credits don't roll over. Top-up packs live longer. On a hobby project this is trivia; in production it is the difference between a plan that funds your pipeline and one that only funds your browser tab.
Net effect: the unlimited offer is an excellent trial, and it is not a cost-of-goods reduction. The discounted surface and the surface you actually pay for are different line items.
Do the math for your own mix
The headline assumes you move everything to T1. You won't, because some of your shots need the other model. So compute the blend.
#!/usr/bin/env python3
"""Blended video-generation cost for a mixed shot list.
RATE: your own effective $/second from your invoice, not a blog post.
MIX: (shot_type, share_of_usable_seconds, model). Shares must sum to 1.0.
"""
RATE = {"t1": 0.22, "seedance": 0.585} # $/s at 1080p, reported 2026-10-24
MIX = [
("dialogue / lip-sync", 0.35, "t1"), # T1 wins lip-sync
("b-roll / product", 0.30, "t1"), # calm camera, low physics risk
("dance / choreography", 0.20, "seedance"), # T1 loses here
("fast action", 0.15, "seedance"), # T1 blurs
]
TAKES = 4 # generations per usable shot — your reroll rate
USABLE_SECONDS = 600 # finished seconds you actually keep per month
generated = USABLE_SECONDS * TAKES
print(f"generated seconds/month: {generated}")
for model in ("t1", "seedance"):
print(f"all-{model:<9} ${generated * RATE[model]:>7,.0f}")
blended = sum(generated * share * RATE[model] for _, share, model in MIX)
solo = generated * RATE["seedance"]
print(f"blended mix ${blended:>7,.0f}")
print(f"saving vs seedance-only: {1 - blended / solo:.0%}")
Output on that mix:
generated seconds/month: 2400
all-t1 $ 528
all-seedance $ 1,404
blended mix $ 835
saving vs seedance-only: 41%
41%, not 62%. That gap is not a trick — it is what routing looks like. And note which way the error runs: 62% is the best case you can only reach by sending every shot to the cheaper model, including the ones it is bad at. The headline number is a ceiling on your saving, never a floor.
Change TAKES to 12 and the absolute numbers triple, but the percentage barely moves — reroll rate scales both sides equally. What actually moves the percentage is the mix, which is why the next table matters more than the pricing page.
A routing table you can steal today
| Shot type | Route to | Why |
|---|---|---|
| Lip-sync dialogue, talking head | T1 | Its clearest win, including non-English |
| Product b-roll, slow push-ins | T1 | Calm camera, no physics to fail |
| Cinematic establishing shots | T1 | Color and camera are its strength |
| Dance, choreography, trend remakes | Seedance 2.5 | Explicitly the weaker model here |
| Fights, falls, collisions | Seedance 2.5 | Fails on contact points and physics |
| Anything with readable on-screen text | Seedance 2.5, or composite the text in post | T1 renders signage illegible |
The last row is worth internalising: if your video needs legible words baked in, that is a compositing problem, not a model problem. Overlay it after generation and you stop paying for a capability neither model reliably has.
A one-hour bench before you switch anything
Five prompts is enough to find the edge, and you should run them on your own footage rather than anyone's highlight reel:
- One calm dialogue shot, face in frame, one line of speech.
- One dance or choreography shot with a full-body subject.
- One fast action shot with a hard stop or impact.
- One product rotation with a reflective surface.
- One shot containing a legible sign or label.
For each, look at the same five things: hands, contact points, text legibility, identity drift across the full duration, and whether the action completes or decays in the last second. Then compare blended cost against the routing table above. You will get roughly the same verdict as the independent test, and you will have your own evidence instead of mine.
What I actually took away
Per-second generation price is converging on zero — T1 at 38% of the official rate, Wan 3.0 marketed as near-Seedance-2.5 at 30% of the price, MiniMax H3 near-Seedance-2.0 at 30% of the cost. Within a year the per-second number will be the least interesting figure in any model comparison.
What stays expensive is everything around it: the rerolls of shots you should not have generated, the latency tax of a shared queue you cannot prioritise, and the work of deciding which shots your product actually needs. A 62% discount on the cheapest line item is welcome. It is not the line item that decides whether the product ships.
So: benchmark your own five prompts, route by shot type, and recompute the blend from your invoice. That gets you most of the discount and none of the regressions.
Sources, so you can check the numbers rather than take them: the launch claim and reach figures come from the launch post as tracked by Doomers; plan pricing and the unlimited-model terms from Topview's own pricing page; the per-second comparison from The AI Dude's review, which flags its pricing as un-re-verified since 23 October; the ten-challenge head-to-head from Art of Smart's writeup of a self-funded comparison; the credit and API wallet rules from TechPilot's reading of Topview's pricing on 18 October. The 62% is Topview's claim. The quality comparison is contested and rests on a small number of independent tests rather than a public benchmark — including mine, which is arithmetic, not measurement.
Top comments (0)