Code2Video-AA Elo
Code · 2026-10-26
Artificial Analysis' implementation of HeyGen's Code2Video Bench: each model writes one HyperFrames HTML composition (HTML layout, GSAP animation) for each of 114 short product and brand video design briefs, rendered to video. A Gemini 4 Argon judge compares two models' videos for the same brief blind on five axes (engagement, prompt intent, composition, temporal consistency, craft); each axis gets an Elo rating and the overall Elo is their mean, anchored to GPT-5.5 (xhigh) at 1000.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Opus 5.5 | 1536 |
| Claude Sonnet 5.5 | 1443 |
| GPT-6 Astra | 1427 |
| GPT-6.1 Sol | 1369 |
| Claude Fable 5.1 | 1325 |
| Gemini 4 Argon | 1259 |
| Claude Haiku 5.5 | 1192 |
| Grok 4.7 | 1127 |
| DeepSeek V4.1 Flash | 1053 |
| MiMo-V2.6-Flash | 1052 |
| MiMo-V2.6-Pro | 1052 |
| Kimi K3 | 1041 |
| GLM-5.3 | 1034 |
| GPT-6 Luna | 1030 |
| Qwen3.8 Max (0902) | 1021 |
| Muse Spark 1.3 | 1019 |
| GPT-5.5 | 1000 |
| Muse Wonder | 997 |
| GLM-5.3 Flash | 975 |
| Qwen3.8 27B | 949 |
| Mistral Large 4 Preview | 931 |
| Gemini 3.8 Flash | 899 |
| Step 5 Preview | 821 |
| MiniMax M3 | 724 |
| Beam 1.0 501B A23B | 675 |
| Inkling | 467 |
| Nemotron 3 Ultra 550B A55B | 304 |