TransitLM - Connectivity: leaderboard
Share (%) of routes from general-purpose LLMs whose consecutive stations are actually connected by a shared line or a valid transfer, on TransitLM (2026): a model must produce a public-transit route between two points in Beijing, Shanghai, Shenzhen or Chengdu from its own knowledge, with no map or routing engine (1,000 test requests). Higher is better.
Metric: Connectivity (%): share of the 1,000 requests whose predicted station sequence has every consecutive pair adjacent on a shared line or linked by a valid transfer, TransitLM Optimal Route Generation, 1,000 test requests across Beijing, Shanghai, Shenzhen and Chengdu given origin and destination names with GPS coordinates and no map or routing engine; general LLMs answer in a lenient format (boarding and alighting station per leg); higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 6 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Setting | Score | Overall rank |
|---|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 75.5 | #99 | |
| 2 | DeepSeek V4 Pro | 64.9 | #196 | |
| 3 | Seed 2.0 Pro | 61.6 | #222 | |
| 4 | GPT-5.4 Pro | xHigh | 60.5 | #91 |
| 5 | Claude Opus 4.6 | 48.1 | #113 | |
| 6 | Qwen 3.6 Plus | 45 | #238 |
No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=transitlm-connectivity · How It Works · Data refreshed daily, snapshot 2026-11-21.