DiagnosticIQ Verbose: leaderboard
Metric: Macro accuracy (%) on DiagnosticIQ Verbose, the core questions with each symbolic rule paraphrased into natural language (mean of the per-asset accuracies), zero-shot multiple choice, temperature 0, at most 4,096 output tokens; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 17 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Opus 4.6 | 71.58 |
| 2 | Claude 3.7 Sonnet | 68.22 |
| 3 | Claude Sonnet 4.6 | 64.81 |
| 4 | O1 | 64.53 |
| 5 | DeepSeek V3 | 62.02 |
| 6 | Qwen 2.5 72B Instruct | 56.73 |
| 7 | Mistral Medium 3 | 55.89 |
| 8 | Gemini 1.5 Pro | 55.78 |
| 9 | Llama 3.3 70B Instruct | 55.16 |
| 10 | Llama 3.1 405B | 53.81 |
| 11 | Mistral Small 3.1 | 53.78 |
| 12 | Granite 3.3 8B Instruct | 53.69 |
| 13 | Gemini 2.0 Flash | 49.78 |
| 14 | Phi-4 | 45.64 |
| 15 | Claude 3.5 Haiku | 42.3 |
Interactive version: theaggregate.ai/benchmark?slug=diagnosticiq-verbose · How It Works · Data refreshed daily, snapshot 2026-10-07.