ClaimDB: A Fact Verification Benchmark over Large Structured Data: leaderboard
ClaimDB evaluates language models on fact verification over large structured databases, where claims must be checked against composed records and multiple tables. The input modality is text. Its pinned metric is accuracy, reflecting the share of claims judged correctly in the source-reported leaderboard. Source: BenchmarkList’s attributed result table.
Metric: Accuracy. Source: benchmarklist.com. Saturation forecast: Estimated already saturated. 30 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 Mini | 0.83 | #193 |
| 2 | Claude Haiku 4.5 | 0.81 | #287 |
| 3 | Gemini 3 Flash | 0.8 | #103 |
| 4 | Gemini 2.5 Flash | 0.79 | #255 |
| 5 | GPT-5 Nano | 0.79 | #439 |
| 6 | GPT-OSS-20B | 0.74 | #523 |
| 7 | Claude 3.5 Haiku | 0.68 | #571 |
| 8 | Ministral 3 14B | 0.62 | #664 |
| 9 | Devstral Small 2 | 0.59 | #491 |
| 10 | Qwen 3 32B | 0.57 | #449 |
| 11 | Ministral 3 8B | 0.53 | #699 |
| 12 | Qwen 3 14B | 0.53 | #542 |
| 13 | QwQ-32B | 0.5 | #422 |
| 14 | Claude 3 Haiku | 0.49 | #801 |
| 15 | GPT-4o Mini | 0.45 | #598 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=claimdb-a-fact-verification-benchmark-over-large-structured-data · How It Works · Data refreshed daily, snapshot 2026-11-05.