LLM-WikiRace — leaderboard
Planning benchmark where models navigate Wikipedia hyperlink graphs from a start page to a target page, testing strategy over real-world knowledge graphs.
Metric: Average Success (%). Source: llmwikirace.github.io. Status: saturation imminent. 29 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 64.93 |
| 2 | GPT-5 | 55.83 |
| 3 | Claude Opus 4.5 | 55.17 |
| 4 | Claude Opus 4.6 | 54.57 |
| 5 | DeepSeek R1 | 54.23 |
| 6 | Gemini 2.5 Flash | 51.4 |
| 7 | GPT-5.2 | 50.23 |
| 8 | GPT-5 Mini | 47.5 |
| 9 | Claude Sonnet 4.5 | 47.3 |
| 10 | Kimi K2 | 46.6 |
| 11 | Grok 4.1 Fast | 46.57 |
| 12 | Gemini 2.0 Flash | 45.1 |
| 13 | Llama 3 70B | 43.6 |
| 14 | Gemma 3 27B | 36.67 |
| 15 | GPT-5 Nano | 33.4 |
Interactive version: theaggregate.ai/benchmark?slug=llm-wikirace · How the rankings work · Data refreshed daily, snapshot 2026-07-22.