Agnes-Video-V2.0 Lands at $0.30/min: A Price Disruptor on the Artificial Analysis Leaderboard
Agnes AI's V2.0 video model dropped this week with native audio-video generation at $0.30/min — roughly 28x cheaper than Seedance 2.0. It's in Artificial Analysis's top 10 by Elo, but well below the top-3 leaders. Here's the honest read on where it fits.
A new entrant landed on the Artificial Analysis video leaderboard this week with a pricing claim sharp enough to warrant a closer look: Agnes-Video-V2.0 at $0.30 per minute with native audio-video generation. To put that number in immediate context — that’s roughly 28× cheaper per second than Seedance 2.0 standard and far below any model in the previous top-10. Agnes AI is openly framing this as the “price butcher” play.
The honest read is more nuanced than the headline. Agnes V2.0 is a meaningful price disruptor. It’s also clearly positioned below the current quality leaders by Elo. Below is what shipped, where Agnes V2 actually sits on the leaderboard, and which workloads the $0.30/min price tag changes.
What shipped
| Detail | Value |
|---|---|
| Model name | Agnes-Video-V2.0 |
| Maker | Agnes AI (AgnesAI Sapiens) |
| Pricing | $0.30/min (~$0.005/sec) |
| Native audio-video | Yes |
| Modes | Text-to-video, image-to-video, first-frame, first-and-last-frame, multi-frame |
| Distribution | Agnes app, PAVO app, API |
| Release timing | Week of May 19, 2026 |
The mode list is worth pausing on. Most video models in this price range ship text-to-video only; Agnes V2.0 supports first-and-last-frame interpolation and multi-frame conditioning natively. That’s a closer feature surface to Seedance 2.0’s standard endpoint than to the cheap-tier Veo or Sora variants.
The leaderboard position, honestly
Agnes’s own announcement says “top 10 AI labs globally.” That framing is accurate but obscures the gap. Pulling the Artificial Analysis Elo numbers Agnes posted alongside the launch:
| Category | Agnes-V2 Elo | Sample size | 95% CI |
|---|---|---|---|
| Text-to-video (with audio) | 885 | 2,463 | ±13 |
| Image-to-video (with audio) | 934 | 2,576 | ±12 |
For context — the same leaderboard’s top entries when I last covered it:
| Model | T2V Elo | Status |
|---|---|---|
| HappyHorse-1.0 | ~1,333 | Mystery model, no API |
| Seedance 2.0 | ~1,273 | Production, live on WaveSpeedAI |
| SkyReels V4 | ~1,245 | Production |
| Kling 3.0 Pro | ~1,241 | Production |
| Agnes-Video-V2.0 | ~885 | Just shipped |
So Agnes V2 is genuinely in the top 10 — but it’s roughly 400 Elo points behind the leaders. A 400-point Elo gap means Agnes loses head-to-head matchups against Seedance 2.0 the overwhelming majority of the time. That’s not a marginal quality difference; it’s a different tier.
The honest framing: Agnes V2.0 isn’t competing with Seedance 2.0 on quality. It’s competing with whatever you’d otherwise pay $0.30 to generate — which, today, is mostly nothing in the frontier tier.
Where the $0.30/min price actually changes the math
The interesting workloads aren’t the ones where Seedance 2.0 is already producing acceptable output. Those will keep using Seedance. The interesting categories are the ones that couldn’t economically run frontier video at all before today:
- High-volume rough draft generation. When unit cost drops from ~$8/min (Seedance 2.0) to $0.30/min, you can generate 25+ variants per dollar where you’d previously generate one. Even if quality drops materially, the prompt-engineering loop accelerates dramatically.
- Personalized content at scale. Per-user video generation for educational, training, or social-feed personalization workflows. The economics that didn’t work at $0.14/sec start to work at $0.005/sec.
- Internal eval pipelines. Generating reference clips to score against new model versions has been a real cost in evaluation-heavy stacks. Dropping reference-generation cost 28× changes how you can structure benchmark loops.
- Markets where premium video is impractical. Long-form educational platforms, low-CPM ad networks, low-budget creator tools — pricing in regions where Seedance/Veo were never feasible.
The category Agnes V2 doesn’t serve is the one most production video workflows care about today: a single high-quality clip that ships to a paying customer. For that, the 400-Elo gap matters. For volume work upstream of that, the gap matters much less.
What we don’t know yet
Three things that aren’t public and would change the read:
- Independent benchmarks. Agnes’s leaderboard ranking is on Artificial Analysis specifically. Whether the same quality holds on different rubrics — text-rendering accuracy, multi-character consistency, contact physics — isn’t documented yet. The “top 10 globally” claim is real on one benchmark; replication would make it real generally.
- Audio-sync quality. Native audio-video is in the spec sheet. Whether the lip-sync, ambient-audio, and music-cue quality match the visual tier isn’t documented. With Veo 3.1 setting a high bar on audio-visual coherence, this is the axis that determines whether “with audio” is a real product or a checkbox.
- Latency and throughput. A $0.30/min model that takes 10 minutes per generation has a very different economics than one that runs in 30 seconds. Artificial Analysis’s “generation time” data wasn’t populated for Agnes V2 when I checked the dashboard.
These won’t resolve from press coverage. They resolve when builders run their own evals, which is the next two to four weeks of public discourse.
Where Agnes V2 fits in the lineup today
Concrete deployment reads:
Use Agnes V2 for:
- Volume A/B testing of prompts where unit cost dominates quality decisions
- Rough-draft generation in storyboarding workflows
- Personalized video at scales that don’t economically work on Seedance/Veo
- Internal eval reference-clip generation
Stay on Seedance / Veo / Sora for:
- Customer-facing final-output video
- Content where the 400-Elo quality gap visibly degrades the result
- Workflows requiring the multi-modal reference inputs Seedance 2.0 specializes in (images + videos + audio combined, as covered in our Seedance 2.0 guide)
- Anything where Artificial Analysis’s “audio-on” Elo gap of 350+ points would be perceptible to the end user
What to watch for
Two signals over the next month:
- Third-party benchmark replication. Does Agnes V2 hold its top-10 position on benchmark suites that aren’t Artificial Analysis? Watch for Vidu cross-platform evals and human-preference panels run by independent labs.
- The next Agnes version. Agnes AI has been iterating quickly — V1.2 → V2.0 happened in a relatively short window. If V2.1 or V3 closes the Elo gap while holding the $0.30/min price, the product story becomes much more interesting. If they raise prices to chase quality, they become a normal mid-tier player.
Until then
For production video workloads today, the Seedance 2.0 lineup — Standard, Fast, the upcoming Mini and 2.1 variants — remains the most defensible default. Agnes V2 is a worth-evaluating addition to the volume-tier of any video stack, not a Seedance replacement. Run your own Elo-style A/B comparison on your specific prompt set before deciding which tier deserves what work.
Sources: Agnes AI’s X announcement, Sohu coverage on the launch, Artificial Analysis Agnes Video page, Text-to-Video leaderboard.