Track and compare AI video generation models across leaderboard position, availability, and pricing with a packaged Claude Code / Cowork skill.
This repository turns ai-video-tracker.skill into a real open-source project: a readable SKILL.md, a reusable model snapshot, and a packaged installer you can upload directly into Claude or inspect and fork yourself.
AI video changes too quickly for static blog posts. A model can top a blind benchmark, disappear from public access, or shift pricing before most people notice. This skill helps answer the practical questions:
- Which models rank well?
- Which ones can I actually use today?
- What do they roughly cost?
- What is the story behind unusual models such as HappyHorse 1.0?
Repository snapshot for v1.0.0, captured on April 9, 2026 from the project's tracked dataset. This is a release snapshot, not a promise that the live leaderboard still looks identical.
| Rank | Model | Creator | Arena Signal | Availability | Rough Pricing | Notes |
|---|---|---|---|---|---|---|
| 1 | HappyHorse 1.0 | Anonymous / disputed | T2V and I2V leader in tracked April 2026 notes | Not publicly available | N/A | High benchmark mystique; detailed analysis |
| 2 | Seedance 2.0 | ByteDance / Dreamina | Strong blind Elo performer | Web UI and API ecosystem | Varies by provider | Best "usable now" option in the bundled snapshot |
| 3 | Kling 3.0 | Kuaishou | Near-top production model | API and web UI | Paid API | Reliable production choice for human motion |
| 4 | Veo 2 | Google DeepMind | Strong technical quality | Google AI Studio / Vertex AI | Enterprise-style pricing | Good fit for Google-heavy teams |
| 5 | Wan 2.1 | Alibaba | Open-source alternative | Weights available | Self-host cost | Best for teams that want control |
For the live leaderboard, verify directly against Artificial Analysis Video Arena. For ongoing HappyHorse-specific coverage, use happyhorse1.info.
Option 1: install from source.
mkdir -p ~/.claude/skills/ai-video-tracker
cp /path/to/this/repo/SKILL.md ~/.claude/skills/ai-video-tracker/SKILL.md
cp -R /path/to/this/repo/data ~/.claude/skills/ai-video-tracker/dataOption 2: use the packaged asset from Releases and unzip it so the archive root is ai-video-tracker/, matching Claude's expected skill-folder structure.
Anthropic's skill docs: Agent Skills
- Download
ai-video-tracker.skillfrom this repository or the latest Release. - Open
Customize > Skills. - Choose
+ Create skill, thenUpload a skill. - Select the packaged
.skillfile.
Anthropic's upload flow: Use Skills in Claude
Prompt: "I need to make a 30-second SaaS product demo video. Which AI video generator should I use if I need something available today?"
Expected direction: Recommend Seedance or Kling first, explain why HappyHorse is interesting but inaccessible, and link to the live tracker for future changes.
Prompt: "What is HappyHorse 1.0 and why do people keep talking about it?"
Expected direction: Explain the benchmark story, the lack of public access, the uncertainty around provenance, and point to happyhorse1.info for continuing updates.
Prompt: "Compare the top AI video models right now with rankings, access, and pricing."
Expected direction: Produce a quick-scan table with benchmark signal, availability, and rough cost so the user can decide quickly.
The bundled dataset lives in data/models.json. It is designed to be easy to fork and update when rankings or product availability move.
Primary sources:
Pull requests are welcome, especially for:
- Model availability updates
- Pricing corrections
- New leaderboard entrants
- Better source attribution in
data/models.json
If you update the snapshot, please include the date and source link in the JSON so others can audit the change.
Want the ongoing version instead of the packaged snapshot? Follow the public tracker at happyhorse1.info.