Code2Video
We show this table for reference; we do not rank on it.
HeyGen's Code2Video tests whether a model can turn a detailed creative brief into a finished motion-graphics video by writing code.
Evaluation results published by HeyGen. How Kaggle Benchmarks reruns tests.
Rating on Code2Video v1 — Latest evaluation October 22, 2026
We mirror the published rating view for Code2Video v1. Claude Opus 5.5 has the highest reported value at 1585. A rating gap smaller than the two rows' intervals is not a reliable difference. We do not use these results to rank models overall.
Claude Opus 5.5
±9; Raw Score 33.5%; Evaluated 2026-10-22
GPT-5.5
±12; Raw Score 34.3%; Evaluated 2026-10-22
GPT-6 Astra
±10; Raw Score 28.9%; Evaluated 2026-10-22
19 source rowsAgenticCurrentDisplay onlyUpdated Latest evaluation October 22, 2026
Rating table (19 source rows)
| Rank | Model / configuration | Score | Parameters (B) | Open / closed |
|---|---|---|---|---|
| 1 | Claude Opus 5.5Creator not reported±9; Raw Score 33.5%; Evaluated 2026-10-22 | 1585 | Not reported | Unknown |
| 2 | GPT-5.5Creator not reported±12; Raw Score 34.3%; Evaluated 2026-10-22 | 1571 | Not reported | Unknown |
| 3 | GPT-6 AstraCreator not reported±10; Raw Score 28.9%; Evaluated 2026-10-22 | 1565 | Not reported | Unknown |
| 4 | GPT-5.6 SolCreator not reported±10; Raw Score 23.3%; Evaluated 2026-10-22 | 1548 | Not reported | Unknown |
| 5 | GPT-6 SolCreator not reported±10; Raw Score 25.1%; Evaluated 2026-10-22 | 1546 | Not reported | Unknown |
| 6 | Claude Opus 5Creator not reported±10; Raw Score 25.7%; Evaluated 2026-10-22 | 1543 | Not reported | Unknown |
| 7 | Gemini 4 ArgonCreator not reported±14; Raw Score 30.9%; Evaluated 2026-10-22 | 1539 | Not reported | Unknown |
| 8 | Claude Fable 5.1Creator not reported±8; Raw Score 19.6%; Evaluated 2026-10-22 | 1536 | Not reported | Unknown |
| 9 | Qwen 3.8 MaxCreator not reported±11; Raw Score 20.7%; Evaluated 2026-10-22 | 1533 | Not reported | Unknown |
| 10 | GLM-5.3Creator not reported±12; Raw Score 14.4%; Evaluated 2026-10-22 | 1496 | Not reported | Unknown |
| 11 | Kimi K3Creator not reported±11; Raw Score 10.4%; Evaluated 2026-10-22 | 1481 | Not reported | Unknown |
| 12 | Gemini 3.7 FlashCreator not reported±13; Raw Score 11.8%; Evaluated 2026-10-22 | 1477 | Not reported | Unknown |
| 13 | Deepseek V4 Flash 0731Creator not reported±11; Raw Score 9.4%; Evaluated 2026-10-22 | 1472 | Not reported | Unknown |
| 14 | GPT-6 LunaCreator not reported±11; Raw Score 8.0%; Evaluated 2026-10-22 | 1469 | Not reported | Unknown |
| 15 | GLM-5.3-FlashCreator not reported±14; Raw Score 8.9%; Evaluated 2026-10-22 | 1467 | Not reported | Unknown |
| 16 | Gemini 3.1 Pro PreviewCreator not reported±13; Raw Score 8.3%; Evaluated 2026-10-22 | 1454 | Not reported | Unknown |
| 17 | Grok 4.6Creator not reported±10; Raw Score 6.9%; Evaluated 2026-10-22 | 1450 | Not reported | Unknown |
| 18 | Gemini 3.8 FlashCreator not reported±13; Raw Score 6.6%; Evaluated 2026-10-22 | 1440 | Not reported | Unknown |
| 19 | Deepseek V4 ProCreator not reported±16; Raw Score 3.9%; Evaluated 2026-10-22 | 1387 | Not reported | Unknown |
Code2Video Arena
Ratings are shown as published, rounded to whole points. The vote count, interval and reasoning setting stay on each row. Missing results are not zero. How Kaggle Benchmarks reruns tests.
| Source rank | Source model or configuration | Arena rating | Source setup | Parameters (B) | Open / closed |
|---|---|---|---|---|---|
| 1 | Gemini 4 Carbon | 1599 | High Reasoning; 93 votes; ±116; Evaluated 2026-11-12Original result | Not reported | Unknown |
| 2 | Claude Opus 5.5 | 1593 | High Reasoning; 713 votes; ±98; Evaluated 2026-11-12Original result | Not reported | Unknown |
| 3 | GPT-6 Astra | 1556 | High Reasoning; 690 votes; ±98; Evaluated 2026-11-12Original result | Not reported | Unknown |
| 4 | Gemini 4 Argon | 1548 | High Reasoning; 734 votes; ±98; Evaluated 2026-11-12Original result | Not reported | Unknown |
| 5 | Claude Sonnet 5.5 | 1542 | High Reasoning; 737 votes; ±97; Evaluated 2026-11-12Original result | Not reported | Unknown |
| 6 | Claude Fable 5.1 | 1526 | High Reasoning; 706 votes; ±98; Evaluated 2026-11-12Original result | Not reported | Unknown |
| 7 | GPT-6.1 Sol | 1512 | High Reasoning; 707 votes; ±98; Evaluated 2026-11-12Original result | Not reported | Unknown |
| 8 | Qwen 3.8 Max | 1476 | Xhigh Reasoning; 597 votes; ±98; Evaluated 2026-11-12Original result | Not reported | Unknown |
| 9 | Claude Haiku 5.5 | 1475 | High Reasoning; 712 votes; ±98; Evaluated 2026-11-12Original result | Not reported | Unknown |
| 10 | GLM-5.3 | 1449 | High Reasoning; 742 votes; ±97; Evaluated 2026-11-12Original result | Not reported | Unknown |
| 11 | Kimi K3 | 1433 | High Reasoning; 705 votes; ±98; Evaluated 2026-11-12Original result | Not reported | Unknown |
| 12 | Gemini 3.8 Flash | 1413 | High Reasoning; 729 votes; ±98; Evaluated 2026-11-12Original result | Not reported | Unknown |
| 13 | GLM-5.3-Flash | 1377 | High Reasoning; 727 votes; ±98; Evaluated 2026-11-12Original result | Not reported | Unknown |
How to read this leaderboard
Use it to compare models on motion design and video production through code. The rating orders the models on this board; the raw score on each row shows how far every model still is from the reference.
Operator receipt: 19 sourced rows are currently displayable on this page; the highest published score among these 19 source rows is Claude Opus 5.5 at 1585, which does not establish a market leader.
Honest limit: The rating is relative to this pool of models and is not a percentage. Results are tied to the Hyperframes environment. The Code2Video Arena table on this page is a separate human-vote board and the two are not interchangeable.
How to read Code2Video
A system receives a creative brief and a video environment built on Hyperframes, and must render a sequence that follows the brief: composition, typography, colour, timing, animation and transitions. The benchmark measures how closely the result executes the brief against a human-made reference, not whether the video is subjectively good. Kaggle publishes a relative rating with an interval and a Raw Score percentage for each model.
The rating is relative to this pool of models and is not a percentage. Results are tied to the Hyperframes environment. The Code2Video Arena table on this page is a separate human-vote board and the two are not interchangeable.
Snapshot
A rating gap smaller than the two rows' intervals is not a reliable difference. The table retains each source configuration and its reported value.
19 source rows are shown for Code2Video. The benchmark falls in the Agentic category. We keep external benchmark mirrors separate from the weighted global scoring system, so these results remain source-specific evidence. Code2Video is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.
About Code2Video
Year
2026
Tasks
Creative briefs rendered as motion-graphics video
Format
Relative rating plus a raw score against human references
Difficulty
Code-driven motion design
A system receives a creative brief and a video environment built on Hyperframes, and must render a sequence that follows the brief: composition, typography, colour, timing, animation and transitions. The benchmark measures how closely the result executes the brief against a human-made reference, not whether the video is subjectively good. Kaggle publishes a relative rating with an interval and a Raw Score percentage for each model. These results are display-only and do not enter overall or category rankings.
Freshness and provenance
Version
Code2Video 2026
Refresh cadence
Quarterly
Staleness state
Current
Question availability
Public benchmark set
BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.
Questions
What does Code2Video measure?
HeyGen's Code2Video tests whether a model can turn a detailed creative brief into a finished motion-graphics video by writing code.
Which source row has the highest reported Code2Video value?
Claude Opus 5.5 has the highest reported Code2Video value at 1585. A rating gap smaller than the two rows' intervals is not a reliable difference. BenchLM shows this benchmark for display only and does not use it in overall rankings.
How many models are evaluated on Code2Video?
The Latest evaluation October 22, 2026 snapshot contains 19 AI models.
Know when it’s worth switching models
The model to choose, the cheaper alternative, and the release we would wait on.
Read a sample issueJoin 7,000+ subscribers.
One email each week. Unsubscribe anytime.