Transparency
What AI clip tools actually cost to run — and why we can price flat
15 September 2026
A note before you read this: the cost and margin figures below are our estimates, built from public cloud pricing for equivalent AI workloads — not numbers StreamLadder, OpusClip, Eklipse, or StreamGen have disclosed. We don’t know their actual infra costs, their negotiated rates, or their internal margins. Our own number is built the same way — from public market rates, not a measured production bill — so we’re holding ourselves to the same estimate methodology we’re applying to everyone else. We’re not naming our exact model/vendor stack here; the point is the cost shape, not the shopping list.
Every AI clip tool, ours included, runs the same two expensive steps on your footage: a transcription pass (turning audio into timestamped text) and an LLM ranking pass (finding which moments are worth clipping). Both scale with the number of minutes you feed in. That’s why almost every competitor in this space metes you with credits, or gates 4K/no-watermark/multi-seat behind a higher tier — the more you use it, the more it costs them to run, so that scaling gets passed straight to your bill.
The public pricing, recapped
All of this is sourced from each vendor’s own pricing page — see /compare for the full table and links.
| Tool | Entry paid tier | Pricing model |
|---|---|---|
| OpenStreamTools | €10/mo | Flat, unlimited |
| StreamLadder | ~$9–11/mo | Tiered, export limits |
| OpusClip | $15/mo (150 credits) | Credit-based, 1 credit/source minute |
| Eklipse | $19.99/mo | Tiered, 720p on entry tier |
| StreamGen | €5.99/mo (Plus) | Tiered, Autopilot on top tier only |
Estimating the AI cost per user
We built this bottom-up from three public, sourced numbers, and one assumption we’re stating outright: an active user processes about 25 hours of VOD a month. Every figure below scales linearly with that assumption — halve the usage, halve the cost.
1. Transcription. Commercial speech-to-text APIs in this class span a real range depending on which provider you pick:
| Provider tier | Rate | Cost at 25 hrs/mo |
|---|---|---|
| Fast/cheap (e.g. Groq’s Whisper-large-v3-turbo) | $0.04/hr | $1.00 |
| Mid (e.g. GPT-4o-mini transcribe) | $0.18/hr | $4.50 |
| Premium (e.g. OpenAI’s Whisper-1) | $0.36/hr | $9.00 |
That’s a 9x spread driven entirely by which transcription backend you pick — and it’s the single biggest lever in this whole cost model.
Transcription cost per hour, by backend tier
2. LLM ranking. A cheap frontier-lite model class on OpenRouter — the kind of model built for exactly this job, not a flagship reasoning model — prices around $0.25 per million input tokens and $1.50 per million output tokens. A VOD hour’s transcript plus prompt overhead runs roughly 20,000 input tokens and the ranked-clip JSON output around 1,500 tokens, so:
(20,000 / 1,000,000 × $0.25) + (1,500 / 1,000,000 × $1.50) ≈ $0.0073 per VOD hour
× 25 hours/month ≈ $0.18/monthRanking is almost a rounding error next to transcription — it’s the per-minute audio processing that decides your unit economics, not the ranking pass.
3. Infra. Cloudflare Workers’ paid plan is a flat $5/month per account (not per user) plus $0.02 per million additional CPU-ms beyond the included 30M/month, and R2 storage runs $0.20/GB-month. Because source video never touches our servers — it’s edited client-side and never uploaded — there’s no video storage line at all, just transcripts and small metadata blobs. Spread the $5 base fee across an active user base of any real size and this rounds to $0.02–0.05 per user.
Putting it together, at 25 hours/month:
| Scenario | Transcription | Ranking | Infra | Total/user/month |
|---|---|---|---|---|
| Cost-optimized backend | $1.00 | $0.18 | $0.03 | ≈ $1.21 (~€1.12) |
| Premium backend | $9.00 | $0.18 | $0.03 | ≈ $9.21 (~€8.50) |
That 9x range is the whole story: a tool’s margin at any given price is almost entirely a function of which transcription API it picked, not cleverness anywhere else in the stack. We’re not naming which end of that range we sit on — but the pipeline choices in the next section explain why we can price flat regardless.
Why we can still do flat
Two structural choices keep our cost low enough to flatten:
- We built our own clip-ranking pipeline, tuned specifically for finding highlights in Twitch VODs rather than using a general-purpose “summarize this video” prompt. A narrower job is a cheaper job to run well.
- The editor itself costs us nothing to run. Cutting, captioning, and exporting the vertical short happens entirely in your browser via WebCodecs — your source footage never uploads to a server, and we never spin up render compute for it. Every competitor above either transcodes server-side or requires an upload step; we don’t have that cost line at all.
Combined, that’s why 4K export, editor invites, and every AI tool can sit in one €10 plan instead of being rationed across tiers.
The estimated margin picture
Here’s the comparison that actually matters: not the cheapest entry tier, but the tier you’d need to match what’s already included in our €10 plan — 4K export, no watermark. Same $1.21–$9.21/month cost range from above applied to each one:
Same AI cost, very different leftovers
Each bar is the monthly price. The dim nub at the start is our estimated AI cost per user — the same cost, for every tool, since it's the same class of workload. Everything past it is margin.
| Tool | Price at feature parity | Est. AI cost/user | Est. gross margin |
|---|---|---|---|
| OpenStreamTools | €10/mo | $1.21–$9.21 | ~15%–89% |
| StreamLadder | €39/mo (Pro, 4K) | $1.21–$9.21 | ~78%–97% |
| OpusClip | $29/mo (Pro — capped at 300 credits, i.e. 5 hrs) | $0.27–$1.87 † | ~94%–99% |
| Eklipse | $19.99/mo (Premium) | $1.21–$9.21 | ~57%–94% |
| StreamGen | €19/mo (Ultimate) | $1.21–$9.21 | ~55%–94% |
† OpusClip’s Pro credits cap usage at 5 hours of source video a month — far below our 25-hour baseline — so we scaled the cost estimate down to match. A hard usage cap doesn’t just protect their margin, it’s why the margin stays this high even on a lower-priced plan.
StreamLadder’s Pro plan is the clearest case. €39/month for 4K export is 3.9x our price for the same feature — and it’s running on the same class of AI workload we costed out above. Even assuming they picked the most expensive transcription option in that range, our estimate still lands their gross margin around 78%. If they picked the cheap end, it’s closer to 97%. We don’t know which — that’s genuinely their call to make, tiered pricing isn’t wrong — but the AI cost that supposedly justifies a 4K paywall is, by this math, a small fraction of what a 4K tier actually charges for it.
Again — every competitor figure here is our estimate from public pricing and public market infra rates, not their disclosed numbers, and our own number is built the same way. If any of them want to publish real figures, we’ll gladly update this table.
The takeaway
Credit systems aren’t a scam — they exist because AI inference cost is real and scales with use. We just made a different bet: build a narrower, cheaper clip-ranking pipeline, push all editing compute into the browser where it costs us nothing, and pass the savings through as one flat price instead of a tier ladder.