Omni-I2C (HTML) - Parametric Accuracy: leaderboard

Metric: Parametric accuracy S_para (0-100): share of the reference's fine-grained parameters (values, ranges, styles) reproduced in the generated code, checked by a Claude Sonnet 4.5 parser against a fixed checklist, on the 166 HTML items; greedy decoding; higher is better. Source: arxiv.org. Saturation forecast: Around September 2028. 13 models tracked.

Top models

#ModelScoreOverall rank
1Gemini 3 Pro56#77
2Claude Sonnet 4.555.3#138
3GPT-5.155.2#131
4Qwen 3 VL 235B A22B Instruct54.3#264
5Qwen 3 VL 32B Instruct54#276
6Qwen 3 VL 8B Instruct51.6#401
7Qwen 2.5 VL 72B Instruct51.4#364
8Qwen 2.5 VL 7B Instruct45.2#643
9Gemma 3 27B (IT)45#509

No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.

Interactive version: theaggregate.ai/benchmark?slug=omni-i2c-html-parametric-accuracy · How It Works · Data refreshed daily, snapshot 2026-10-11.