OCR-MetaReasoning - RPCS: leaderboard
OCR-MetaReasoning Reasoning Process Compliance Score: mean RPCS over deductive, inductive and abductive items, built from process criteria for capability match, groundedness, step completeness and non-hallucination (arXiv 2608.30678v1, Table 3 AVG. column). Percentage points, higher is better; it scores the reasoning process, not the answer; reasoning modes follow the paper's Table 24.
Metric: Reasoning Process Compliance Score (%). Source: arxiv.org. Saturation forecast: Estimated already saturated. 18 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) (Thinking) | 96.2 | #64 (Gemini 3.1 Pro (Preview)) |
| 2 | GPT-5.4 (Medium) | 95.2 | #83 (GPT-5.4) |
| 3 | GPT-5.4 (Non-reasoning) | 92.5 | #83 (GPT-5.4) |
| 4 | Claude Sonnet 4.6 (Thinking) | 91.9 | #94 (Claude Sonnet 4.6) |
| 5 | Kimi K2.5 (Non-reasoning) | 91.1 | #152 (Kimi K2.5) |
| 6 | Qwen 3 VL 235B A22B (Thinking) | 89.6 | #236 |
| 7 | GPT-5.4 Mini (Medium) | 88.9 | #228 (GPT-5.4 Mini) |
| 8 | Qwen 3 VL 235B A22B Instruct | 87 | #280 |
| 9 | Grok 4.20 (Non-reasoning) | 85.7 | #179 (Grok 4.20) |
| 10 | Qwen 3 VL 30B A3B (Thinking) | 84.4 | #348 |
| 11 | GLM-4.6V (Reasoning) | 84.1 | #330 (GLM-4.6V) |
| 12 | Qwen 3 VL 8B (Thinking) | 82.7 | #423 |
| 13 | GPT-5.4 Mini (Non-reasoning) | 81.4 | #228 (GPT-5.4 Mini) |
| 14 | Qwen 3 VL 30B A3B Instruct | 76.3 | #390 |
| 15 | Qwen 3 VL 8B Instruct | 72.5 | #419 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=ocr-metareasoning-rpcs · How It Works · Data refreshed daily, snapshot 2026-11-05.