AgentDoxx
Harbor: pendingAn evaluation suite of 822 synthetic interview transcripts grounded in public info to measure LLM re-identification risks via web search.
- Samples
- 822
- Released
- 2026
- Scores
- 9
TaoD2C-Bench
TaoD2C-Bench: Benchmarking MLLMs for Industrial UI Code Generation Beyond Visual Fidelity
Evaluates MLLMs on industrial UI code generation using 2,861 designs and 97,652 expert annotations across component, group, alignment, and position constraints.
How to run
ltb · taod2c-bench
$ npm i -g lettokenburn-cli$ ltb bench info taod2c-bench
External harness. No Harbor adapter has been generated for this benchmark yet. `ltb bench info` prints the known links.
Runs happen on your machine with your own subscription; only the score summary is pushed. CLI docs
Scores
8 sourced entries| Harness | Source | Confidence | Link | |||
|---|---|---|---|---|---|---|
| Kimi K3 | 63.3 | N/A | Independent | medium | Oct 7, 2026 | arxiv.org |
| GPT-5.6-terra | 62.5 | N/A | Independent | medium | Oct 7, 2026 | arxiv.org |
| Gemini-3.5-flash | 62.1 | N/A | Independent | medium | Oct 7, 2026 | arxiv.org |
| Qwen3.8 Max | 61.8 | N/A | Independent | medium | Oct 7, 2026 | arxiv.org |
| GPT-5.4 | 60.2 | N/A | Independent | medium | Oct 7, 2026 | arxiv.org |
| Claude Sonnet 5 | 59.9 | N/A | Independent | medium | Oct 7, 2026 | arxiv.org |
| Gemini 3.1 Pro | 57.7 | N/A | Independent | medium | Oct 7, 2026 | arxiv.org |
| Qwen3-VL-235B | 31.6 | N/A | Independent | medium | Oct 7, 2026 | arxiv.org |
Every number links to where it was published. Low-confidence rows are muted and need per-cell verification.
Reproduced by the community
All runsLoading community runs…
About
TaoD2C-Bench comprises 2,861 production UI designs from 17 platforms with expert-annotated implementation requirements. It assesses end-to-end code generation, requirement inference, and realization.
The benchmark uses expert annotations to verify constraint satisfaction rather than just visual fidelity. Experiments reveal significant gaps in MLLM performance, showing visual reconstruction does not guarantee requirement-compliant code.