Code2Video-AA: Claude Opus 5.5 Tops Marketing Video Coding Bench by 93 Elo at 10x the Cost
ArtificialAnlys · x · 2026-10-27
Artificial Analysis launched Code2Video-AA, an independent implementation of HeyGen's Code2Video Bench, testing how well 27 models turn 114 product/brand design briefs into short marketing videos written as HyperFrames HTML compositions (HTML + GSAP, single response, no tools or network). A blind Gemini 4 Argon judge compares rendered videos pairwise across five axes — engagement, prompt intent, composition, temporal consistency, craft — with 35,000+ verdicts converted into per-axis Elo ratings (anchored to GPT-5.5 xhigh at 1000).
Key findings:
- Claude Opus 5.5 (max with fallback) leads at 1536 Elo, ranking first on all five axes, 93 points ahead of the runner-up; Claude Sonnet 5.5 (1443) and GPT-6 Astra (1427) are statistically tied for second.
- Cost per task varies over 300x ($0.01–$3.89): Opus 5.5 costs $3.08/task, while fourth-place GPT-6.1 Sol (1369) costs just $0.31 — about 90% cheaper, trailing by 167 points. Eight models sit on the Elo-vs-cost Pareto frontier.
- Open-weight models trail by at least 483 points: the best (DeepSeek V4.1 Flash, MiMo-V2.6-Flash/Pro, 1052) cost only $0.01–$0.08 per task.
- Craft shows the widest model spread; temporal consistency the narrowest. The judge model itself, Gemini 4 Argon, ranks 6th at 1259.
More from Models
- Fable 5.5 rumored to drop tomorrow, per developer teaser — daniel_mac8 · 2026-10-27
- Enterprise benchmark: GPT 6.1 Sol High scores 94.8% at $1.07 per task, DeepSeek 1017 78.1% at $0.42 — emilahlback · 2026-10-27
- Cohere releases Qwen investigation traces for independent security verification — Cohere · 2026-10-27
- X bundles Grok, Grok Bots and Cursor into one subscription, with Ultra tier at $2,000/year — elonmusk · 2026-10-27
- Simple seed-based method elicits 60 strikingly diverse personas, and Yacine finds it hilarious — yacineMTB · 2026-10-27
- Code2Video-AA Cost Curve: 8 Models on the Pareto Frontier, High Elo Needn't Be Pricey — ArtificialAnlys · 2026-10-27