TKFQA (Table-Text-KG Order): leaderboard
TKFQA (2026) asks counterfactual multi-hop questions whose evidence is spread over a table, text passages and a knowledge graph, so the answer must follow the given context rather than memory. Exact match of the final answer in percent, higher is better, with the contexts given in table, text, knowledge-graph order.
Metric: Exact match (%; final answers to counterfactual multi-hop questions grounded in a table, text passages and a knowledge graph, 1,012 test items; contexts given in table, text, knowledge-graph order; mean of three runs). Source: arxiv.org. Saturation forecast: Estimated already saturated. 14 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | DeepSeek V4 Flash | 84.66 | #254 |
| 2 | O3 | 84.5 | #224 |
| 3 | Gemini 2.5 Flash | 84.2 | #452 |
| 4 | DeepSeek V3.2 | 83.05 | #403 |
| 5 | O4 Mini | 82.5 | #354 |
| 6 | Kimi K2.5 | 79.31 | #252 |
| 7 | Grok 4.3 | 79.22 | #332 |
| 8 | GPT-5 | 75.38 | #163 |
| 9 | MiniMax-M2.7 | 72.42 | #519 |
| 10 | GPT-4.1 Mini | 71.44 | #775 |
| 11 | Gemini 2.5 Flash Lite | 69.93 | #958 |
| 12 | Qwen 3 30B A3B | 64.9 | #1124 |
| 13 | Gemma 3 12B | 62.33 | #2013 |
| 14 | Qwen 3 8B | 52.35 | #1838 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=tkfqa-table-text-kg-order · How It Works · Data refreshed daily, snapshot 2026-11-12.