IdeaAMBIG - Defect Localization - Controlled Synthetic: leaderboard
IdeaAMBIG (arXiv 2609.10539v1, Table 1): the model must name the implementation-critical blocker in an underspecified research idea and assign both taxonomy labels, on controlled synthetic specifications derived from ideation-execution trajectories. Score is Macro Defect Recovery Rate over gold Level-2 defect categories, which needs correct localization and both labels, in percent; higher is better. Zero-shot fixed prompts at temperature 0.
Metric: Macro Defect Recovery Rate (%). Source: arxiv.org. Saturation forecast: Around 2032. 13 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.6 Sol | 12.2 | #22 |
| 2 | Claude Sonnet 5 | 11.4 | #65 |
| 3 | Kimi K3 | 8.2 | #35 |
| 4 | DeepSeek V3.2 | 8 | #210 |
| 5 | Gemini 3.1 Pro (Preview) | 7.6 | #64 |
| 6 | Gemma 4 31B (IT) | 7 | #178 |
| 7 | GLM-5.2 | 7 | #80 |
| 8 | Qwen 3 32B | 5.8 | #449 |
| 9 | Qwen 3.5 9B | 5.5 | #391 |
| 10 | Qwen 3 8B | 4.6 | #678 |
| 11 | Qwen 3.5 397B A17B | 4.6 | #154 |
| 12 | GPT-OSS-120B | 3.3 | #346 |
| 13 | DeepSeek R1 0528 | 2.2 | #232 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=ideaambig-defect-localization-controlled-synthetic · How It Works · Data refreshed daily, snapshot 2026-11-05.