AI Video Model Rankings (June 2026): Who Actually Leads the Leaderboards
Current Artificial Analysis standings decoded: HappyHorse-1.0 tops image-to-video at 1,415 Elo, Kling 3.0 Omni Pro at 1,299, Seedance 2.0 leads the with-audio board — plus Veo 3.1, Runway Gen-4.5, and Hailuo 2.3 compared.
The AI video leaderboard flipped twice in 2026. In February, ByteDance's Seedance 2.0 took the top spot on Artificial Analysis' with-audio board (1,213 Elo). In April, an anonymous model swept every benchmark — and on April 10, Alibaba confirmed it was theirs: HappyHorse-1.0, now leading image-to-video at 1,415 Elo, a full 116 points ahead of Kling 3.0 Omni Pro (1,299). Meanwhile OpenAI exited the consumer market entirely, shutting down Sora on April 26.
This page is our running snapshot of who actually leads — by benchmark board, not marketing copy — and what each leader is best at. We update it as the boards move; figures below reflect June 2026.
The June 2026 standings at a glance
| Model | Maker | Headline benchmark | Native audio | Known for |
|---|---|---|---|---|
| HappyHorse-1.0 | Alibaba (Taotian Future Life Lab) | #1 image-to-video, 1,415 Elo | Yes (7-language lip-sync) | Swept April benchmarks anonymously |
| Kling 3.0 Omni | Kuaishou | i2v 1,299 Elo (Pro) | Yes (5-language lip-sync, voice binding) | 15s clips, multi-shot audio timeline |
| Seedance 2.0 | ByteDance | #1 with-audio board, 1,213 Elo | Yes (8+ language lip-sync) | Unified A/V generation, 12 reference assets |
| Veo 3.1 | Top-tier overall | Yes (most polished integration) | Best all-around output quality | |
| Runway Gen-4.5 + Aleph | Runway | Control benchmarks | Partial | Best control surface; film production ecosystem |
| Hailuo 2.3 | MiniMax | Speed/quality balance | Yes | Fast everyday generation |
Read the boards carefully: Artificial Analysis runs separate leaderboards for text-to-video, image-to-video, and with-audio generation. HappyHorse's 1,415 is an image-to-video figure; Seedance's 1,213 leads the with-audio board. A model can top one board and trail on another — most '#1 AI video model' claims you see online quietly cherry-pick the board.
HappyHorse-1.0 — the April earthquake
An unlabeled model started topping every Artificial Analysis benchmark in early April. On April 10, Alibaba confirmed ownership: HappyHorse-1.0 was built by the Future Life Lab inside Taotian Group, led by Zhang Di — the former Kuaishou VP who ran Kling's technical team before joining Alibaba at the end of 2025. The Kling-beats-Kling subplot is real: its 1,415 i2v Elo sits 116 points above Kling 3.0 Omni Pro. It ships phoneme-level lip-sync in seven languages (English, Mandarin, Cantonese, Japanese, Korean, German, French).
Kling 3.0 Omni — the iteration machine grows audio
Kuaishou's answer to the audio era: up to 15 seconds of continuous video with native audio, lip-sync across five major languages, and a unified multimodal framework that lets you bind a specific voice to a character via video extraction or image-audio pairing. Generation speed remains Kling's trademark. Our hands-on comparison data covers Kling 2.0 vs Seedance 2.0 — the 3.0 Omni matchup is covered in our dedicated head-to-head.
Seedance 2.0 — still the audio-first workhorse
Four months after launch (February 12, 2026), Seedance 2.0 still leads the with-audio leaderboard at 1,213 Elo. Its case: true single-pass audio-video generation, 8+ language lip-sync, up to 12 multimodal reference assets, and 15-second 1080p clips. It is also the cheapest flagship to actually run — see our complete pricing breakdown, with per-second access from ≈$0.03 on Sora2U.
Veo 3.1, Runway Gen-4.5, Hailuo 2.3 — the specialists
- Veo 3.1 (Google) — the most polished audio implementation of any model; the default pick when sound quality on product demos and explainers matters more than price.
- Runway Gen-4.5 + Aleph — the best control surface in the industry and a film-production ecosystem nothing else matches; Aleph adds text-instructed editing of existing footage.
- Hailuo 2.3 (MiniMax) — the speed/quality sweet spot for everyday social content without a steep workflow tax.
What happened to Sora and Sora 2?
OpenAI discontinued the Sora consumer app (web + iOS) on April 26, 2026; the Sora 2 API ($0.10–0.70/sec) follows on September 24, 2026. No replacement product has been announced. Migration paths and data export steps are in our Sora shutdown guide, and the deeper question gets its own article: Is Sora coming back?
Run the with-audio leaderboard champion
Seedance 2.0 — #1 on the Artificial Analysis with-audio board — online at ≈$0.03/sec. Free trial credits, no enterprise verification.
How to choose (60-second version)
- Best raw output + audio polish → Veo 3.1
- Best image-to-video fidelity → HappyHorse-1.0 (where you can access it)
- Best dialogue/lip-sync per yuan → Seedance 2.0
- Best control for film work → Runway Gen-4.5
- Fastest iteration on a budget → Kling 3.0 / Hailuo 2.3
- Full spec-by-spec breakdowns live in our tools comparison hub.
Frequently Asked Questions
What is the best AI video model right now (June 2026)?
It depends on the benchmark board: HappyHorse-1.0 (Alibaba) leads image-to-video at 1,415 Elo on Artificial Analysis; Seedance 2.0 (ByteDance) leads the with-audio board at 1,213 Elo; Veo 3.1 has the most polished audio integration; Runway Gen-4.5 leads on controllability. There is no single #1 across all boards.
What is HappyHorse-1.0 and who built it?
HappyHorse-1.0 is the model that anonymously topped every Artificial Analysis benchmark in April 2026. On April 10, Alibaba confirmed it was built by the Future Life Lab inside its Taotian Group, led by Zhang Di — the former Kuaishou VP who previously ran Kling's technical team. It scores 1,415 Elo on image-to-video and supports phoneme-level lip-sync in seven languages.
Is Kling 3.0 Omni better than Seedance 2.0?
On the image-to-video board, Kling 3.0 Omni Pro (1,299 Elo) ranks higher; on the with-audio board, Seedance 2.0 leads at 1,213. Kling 3.0 offers voice binding and 5-language lip-sync; Seedance counters with 8+ languages, 12 reference assets, and a lower per-second price. See our dedicated Kling 3.0 vs Seedance 2.0 comparison for the full breakdown.
Are these Elo scores reliable?
Artificial Analysis Elo is crowd-judged pairwise preference, which captures perceived quality well but not controllability, speed, or cost. Treat the boards as a quality signal, then weigh workflow factors — clip length, audio, price per second, API access — for your use case.
Which ranked model is cheapest to actually use?
Among the leaders, Seedance 2.0 has the lowest uncapped per-second route — ≈$0.03/sec via Sora2U versus ~$0.14/sec on the official Volcengine API, with free daily credits available through Dreamina and Jimeng. Our Seedance pricing guide compares every channel.