FormInv: leaderboard
A paraphrase-quality audit of MathCheck (ICLR 2025) detected 4 semantically incorrect paraphrases in 129 groups (3.1%); removing them drops GPT-4o from rank 2 to rank 4 and elevates Claude Haiku and DeepSeek V3 above it; these ranking changes are invisible to any single-model evaluation. Cross-model unanimity found these errors automatically (>= 3/4 models for MathCheck; >= 6/9 for our primary eva
Metric: SCR (self-reported). Source: benchmarklist.com. Saturation forecast: Not forecast. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4o | 82 | #343 |
| 2 | DeepSeek V3 | 82 | #321 |
| 3 | Gemini 2.5 Flash | 80 | #245 |
| 4 | O4 Mini | 80 | #180 |
| 5 | GPT-4o Mini | 78 | #599 |
| 6 | DeepSeek R1 | 78 | #251 |
| 7 | Claude Sonnet 4.6 | 74 | #90 |
| 8 | Llama 3.3 70B | 66 | #579 |
| 9 | Claude Haiku 4.5 | 50 | #280 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=forminv · How It Works · Data refreshed daily, snapshot 2026-10-20.