Benchmark Almanac

Code2Video Benchmark

Claude Opus 5.5 1,585 leads since Sep 2026 Not in model cards

168 product-video briefs. Agent writes HTML motion-graphics code rendered by HyperFrames. A trained pairwise judge on five axes produces Elo.

Repo is the HyperFrames renderer, not a task repo. Judge agrees with humans ~82%. Kaggle hosts briefs and runs evaluations.

Made by HeyGen Research + Kaggle Released Sep 2026 Video generation All tasks public Last model added Sep 2026 Scores refreshed 2 Nov 2026
Level
Expert Specialists with years of training, often over hours of work
Scored
Graded by a model A language model scores each response, usually against a rubric.
Answers
Open-ended Free-form text, documents, designs or media judged as a whole.
heygen-com/hyperframes 76,759 stars Paper Leaderboard
19
Models scored on its own leaderboard

Claude Opus 5.5 leads with 1,585

Best score per model from the benchmark's own leaderboard, read on 2 Nov 2026. Where a model ran with several settings or harnesses, we keep its best.

19 models. Scroll sideways for the rest. Bars start at 1,350.

  1. 1,585 Claude Opus 5.5
  2. 1,571 GPT-5.5
  3. 1,565 GPT-6 Astra
  4. 1,548 GPT-5.6 Sol
  5. 1,546 GPT-6 Sol
  6. 1,543 Claude Opus 5
  7. 1,539 Gemini 4 Argon
  8. 1,536 Claude Fable 5.1
  9. 1,533 Qwen 3.8 Max
  10. 1,496 GLM-5.3
  11. 1,481 Kimi K3
  12. 1,477 Gemini 3.7 Flash
  13. 1,472 DeepSeek V4 Flash 0731
  14. 1,469 GPT-6 Luna
  15. 1,467 GLM-5.3-Flash
  16. 1,454 Gemini 3.1 Pro
  17. 1,450 Grok 4.6
  18. 1,440 Gemini 3.8 Flash
  19. 1,387 DeepSeek-V4-Pro

Bars start at 1,350.

  1. 1 Claude Opus 5.5 1,585
  2. 2 GPT-5.5 1,571
  3. 3 GPT-6 Astra 1,565
  4. 4 GPT-5.6 Sol 1,548
  5. 5 GPT-6 Sol 1,546
  6. 6 Claude Opus 5 1,543
  7. 7 Gemini 4 Argon 1,539
  8. 8 Claude Fable 5.1 1,536
  9. 9 Qwen 3.8 Max 1,533
  10. 10 GLM-5.3 1,496
  11. 11 Kimi K3 1,481
  12. 12 Gemini 3.7 Flash 1,477
  13. 13 DeepSeek V4 Flash 0731 1,472
  14. 14 GPT-6 Luna 1,469
  15. 15 GLM-5.3-Flash 1,467
  16. 16 Gemini 3.1 Pro 1,454
  17. 17 Grok 4.6 1,450
  18. 18 Gemini 3.8 Flash 1,440
  19. 19 DeepSeek-V4-Pro 1,387
#ModelScoreSettingDateSource
1 Claude Opus 5.5 1,585 Sep 2026 Official leaderboard
2 GPT-5.5 1,571 Apr 2026 Official leaderboard
3 GPT-6 Astra 1,565 Sep 2026 Official leaderboard
4 GPT-5.6 Sol 1,548 Jul 2026 Official leaderboard
5 GPT-6 Sol 1,546 Sep 2026 Official leaderboard
6 Claude Opus 5 1,543 Jul 2026 Official leaderboard
7 Gemini 4 Argon 1,539 Sep 2026 Official leaderboard
8 Claude Fable 5.1 1,536 Sep 2026 Official leaderboard
9 Qwen 3.8 Max 1,533 Aug 2026 Official leaderboard
10 GLM-5.3 1,496 Aug 2026 Official leaderboard
11 Kimi K3 1,481 Jul 2026 Official leaderboard
12 Gemini 3.7 Flash 1,477 Aug 2026 Official leaderboard
13 DeepSeek V4 Flash 0731 1,472 Jul 2026 Official leaderboard
14 GPT-6 Luna 1,469 Sep 2026 Official leaderboard
15 GLM-5.3-Flash 1,467 Aug 2026 Official leaderboard
16 Gemini 3.1 Pro 1,454 Feb 2026 Official leaderboard
17 Grok 4.6 1,450 Aug 2026 Official leaderboard
18 Gemini 3.8 Flash 1,440 Sep 2026 Official leaderboard
19 DeepSeek-V4-Pro 1,387 Apr 2026 Official leaderboard

2 models have held number one since its release in Sep 2026

Lab claims count from the day the card came out, independent runs from the model’s release.

OpenAI Anthropic
Own leaderboard 2 leaders
Claude Opus 5.5 1,585

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

On the benchmark’s own leaderboard
1,350 1,400 1,450 1,500 1,550 1,600 2026 1,350 1,400 1,450 1,500 1,550 1,600 2026

No lab model card reports it yet

We read the headline results of 479 launch posts from 51 labs, and none of them list this benchmark.

More in Audio and video

All 168 in Audio and video
  • Video-MMMU In use 300 expert-level lecture videos with questions testing perception, comprehension and adaptation of knowledge. Accuracy. Qwen 3.8 Max 88.7%
  • LVBench In use Hours-long video understanding with multiple-choice questions on reasoning, grounding and summarization. Gemini 4 Argon 91.7%
  • MMVU In use 3,000 expert-annotated questions on videos across 27 disciplines. Multiple-choice and open-ended accuracy. Qwen 3.8 Max 82.4%
  • MLVU In use Multi-task long video understanding (minutes to hours) with multiple-choice and generation tasks. Accuracy. Qwen 3.8 Max 90.8%
  • VoiceBench In use Spoken instructions with speaker, environment and content variation. Voice assistant responses scored by GPT judge or rules. Qwen3.5-Omni-Plus 93.1%
  • TVBench In use Models answer multiple-choice questions on 10 temporal video tasks, such as action counting and ordering, that static frames cannot solve. Scored by accuracy. Qwen 3.8 Max 81.9%