Benchmarks · reasoning
GDP.pdf
Can a model answer real professional questions about the PDFs work runs on — manuals, wiring diagrams, filings, forms — so that every rubric criterion an expert wrote is met? Maintained by Surge AI. Score: The board's all-pass rate: the share of responses that satisfy every rubric criterion, judged by Gemini 3.5 Flash, over the 100 held-out tasks and five runs per task.
36 models · 46 results · data as of 2026-10-29 · surgehq.ai ↗
Leaderboard
GPT-6 Astra medium
All results
46 rows| # | Model | Effort | GDP.pdf |
|---|---|---|---|
| 1 | GPT-6 Astra | medium | 35.4% |
| 2 | GPT-6 Astra | max | 34.2% |
| 3 | Opus 5.5 | max | 31.4% |
| 4 | GPT-5.6 Sol | max | 30.7% |
| 5 | Fable 5 | max | 29.8% |
| 6 | Fable 5.1 | max | 29.6% |
| 7 | Muse Spark 1.3 | max | 27.6% |
| 8 | Muse Spark 1.3 | xhigh | 27.6% |
| 9 | GPT-6 Sol | max | 26.4% |
| 10 | GPT-5.5 | xhigh | 26.0% |
| 11 | GPT-6 Sol | medium | 26.0% |
| 12 | GPT-5.6 Terra | medium | 24.7% |
| 13 | Opus 4.8 | max | 24.0% |
| 14 | Opus 5 | max | 24.0% |
| 15 | Gemini 3.7 Flash | high | 23.8% |
| 16 | Gemini 3.8 Flash | medium | 23.4% |
| 17 | Gemini 3.8 Flash | high | 23.2% |
| 18 | Qwen 3.8 Max | xhigh | 23.2% |
| 19 | GPT-6 Luna | max | 23.0% |
| 20 | Grok 4.7 | xhigh | 22.8% |
| 21 | GPT-5.6 Luna | medium | 22.7% |
| 22 | Gemini 3.7 Flash | medium | 21.8% |
| 23 | Grok 4.7 | high | 21.2% |
| 24 | Opus 4.7 | max | 21.0% |
| 25 | GPT-6 Luna | medium | 20.4% |
| 26 | DeepSeek V4.1 Flash | max | 19.8% |
| 27 | Kimi K3 | max | 19.0% |
| 28 | DeepSeek V4.1 Flash | high | 18.2% |
| 29 | Sonnet 4.6 | max | 18.0% |
| 30 | Grok 4.6 | xhigh | 17.2% |