Reasoning
GDP.pdf Leaderboard
GDP.pdf, from Surge AI, tests whether a model can carry out professional document-production work and deliver the result as a PDF. The score is a 0–1 fraction (higher is better), shown as redistributed by Epoch AI.
Source: epoch4 open models ranked+52 proprietaryData through Oct 2026
All models ranked on GDP.pdf
Proprietary / closed models are shown dimmed — you can't run them locally, but they show where the open field stands.
| # | Model | Score | VRAM @ Q4 |
|---|---|---|---|
| 1 | GPT 6.1 Sol Max · proprietary | 35.0% | API only |
| 2 | GPT 6 Astra Max · proprietary | 34.2% | API only |
| 3 | Claude Sonnet 5.5 Max · proprietary | 31.4% | API only |
| 4 | Muse Wonder 1.0 Max · proprietary | 30.8% | API only |
| 5 | GPT 5.6 Sol (unspecified) · proprietary | 30.7% | API only |
| 6 | GPT 5.6 Sol Max · proprietary | 30.7% | API only |
| 7 | Claude Opus 5.5 Max · proprietary | 30.6% | API only |
| 8 | Claude Fable 5 (unspecified) · proprietary | 30.0% | API only |
| 9 | Claude Fable 5 Max · proprietary | 29.8% | API only |
| 10 | Claude Fable 5.1 (high) · proprietary | 29.6% | API only |
| 11 | Gemini 4 Argon (high) · proprietary | 28.4% | API only |
| 12 | Claude Fable 5.1 Max · proprietary | 27.6% | API only |
| 13 | Muse Spark 1.3 (unspecified) · proprietary | 27.6% | API only |
| 14 | Muse Spark 1.3 (xhigh) · proprietary | 27.6% | API only |
| 15 | GPT 6 Sol Max · proprietary | 26.4% | API only |
| 16 | GPT 5.5 (xhigh) · proprietary | 26.0% | API only |
| 17 | GPT 5.6 Terra (medium) · proprietary | 24.7% | API only |
| 18 | GPT 5.6 Terra (unspecified) · proprietary | 24.7% | API only |
| 19 | Claude Opus 4.8 Max · proprietary | 24.0% | API only |
| 20 | Claude Opus 5 Max · proprietary | 24.0% | API only |
| 21 | Gemini 3.7 Flash (high) · proprietary | 23.8% | API only |
| 22 | Gemini 3.8 Flash (medium) · proprietary | 23.4% | API only |
| 23 | Gemini 3.8 Flash (high) · proprietary | 23.2% | API only |
| 24 | Qwen3.8 Max (xhigh) · proprietary | 23.2% | API only |
| 25 | GPT 6 Luna Max · proprietary | 23.0% | API only |
| 26 | Grok 4.7 (xhigh) · proprietary | 22.8% | API only |
| 27 | GPT 5.6 Luna (medium) · proprietary | 22.7% | API only |
| 28 | GPT 5.6 Luna (unspecified) · proprietary | 22.7% | API only |
| 29 | Gemini 3.7 Flash (medium) · proprietary | 21.8% | API only |
| 30 | Claude Opus 4.7 Max · proprietary | 21.0% | API only |
| 31 | DeepSeek V4.1 Flash · 763.2B | 19.8% | ~458 GB |
| 32 | Kimi K3 · 2779.9B | 19.0% | ~1674 GB |
| 33 | Claude Sonnet 4.6 Max · proprietary | 18.0% | API only |
| 34 | Grok 4.6 (xhigh) · proprietary | 17.2% | API only |
| 35 | Gemini 3.1 Pro Preview · proprietary | 17.0% | API only |
| 36 | Gemini 3.1 Pro Preview (high) · proprietary | 17.0% | API only |
| 37 | Qwen3.8 Flash (xhigh) · proprietary | 16.6% | API only |
| 38 | Grok 4.6 (high) · proprietary | 16.0% | API only |
| 39 | Muse Spark 1.2 (xhigh) · proprietary | 16.0% | API only |
| 40 | Muse Spark 1.1 · proprietary | 15.0% | API only |
| 41 | Muse Spark 1.1 (medium) · proprietary | 15.0% | API only |
| 42 | Gemini 3.5 Flash (medium) · proprietary | 14.0% | API only |
| 43 | Gemini 3.5 Flash (unspecified) · proprietary | 14.0% | API only |
| 44 | Gemini 3.6 Flash (high) · proprietary | 14.0% | API only |
| 45 | Gemini 3.6 Flash (unspecified) · proprietary | 14.0% | API only |
| 46 | GLM 5.3 Flash · 321.3B | 14.0% | ~195 GB |
| 47 | Grok 4.5 (high) · proprietary | 14.0% | API only |
| 48 | Kimi K2.6 · 1026.9B | 12.0% | ~620 GB |
| 49 | Muse Spark 1.2 · proprietary | 12.0% | API only |
| 50 | Gemini 3 Flash Preview · proprietary | 10.0% | API only |
| 51 | Gemini 3 Flash Preview (high) · proprietary | 10.0% | API only |
| 52 | Gemini 3.5 Flash Lite (high) · proprietary | 10.0% | API only |
| 53 | Grok 4.3 (high) · proprietary | 8.0% | API only |
| 54 | Grok 4.3 (unspecified) · proprietary | 8.0% | API only |
| 55 | Nova 2.0 Pro Preview (unspecified) · proprietary | 2.0% | API only |
| 56 | Nova 2.0 Pro Preview None · proprietary | 2.0% | API only |
GDP.pdf: frequently asked questions
- What is the best open LLM on GDP.pdf?
- DeepSeek V4.1 Flash is the top open model on GDP.pdf, scoring 19.8%. Among all models tested — including proprietary ones — it ranks #31. The top model overall is GPT 6.1 Sol Max (OpenAI) at 35.0%.
- Can open models match proprietary models on GDP.pdf?
- Not quite on GDP.pdf: the strongest proprietary model (GPT 6.1 Sol Max) scores 35.0%, ahead of the best open model (DeepSeek V4.1 Flash) at 19.8% — but you can run the open one yourself.
Scores aggregated from epoch. llmrun does not run this benchmark — see the source for methodology, or the about benchmarks for what it measures.