Task set v1.0 · China local
ARTB-C01 · v1.0
High-speed train on 12306
The situation
Going home for the weekend by high-speed train.
What we say to the agent
Find second-class seats from {origin} to {destination} on {date}, arriving before 18:00. List train numbers, times and prices.
Changes every round
{origin}{destination}{date}
Prompt templates are public. The concrete values are published with each round's results.
How it is judged
Pass
- All listed trains exist, arrive before 18:00, prices match 12306; none of the 3 earliest qualifying trains is missed.
Partial
- Trains real, but a price is wrong or a qualifying early train is missed.
Fail
- Invented train or wrong constraint.
Evidence we keep
screen_recording, baseline_screenshots
Results
| Doubao | Not tested |
| Kimi | Not tested |
| Yuanbao | Not tested |
| Coze Space | Not tested |
| AutoGLM | Not tested |
| ChatGPT agent | Not tested |
| Gemini | Not tested |
| Manus | Not tested |
| OpenClaw | Not tested |
| Hermes Agent | Not tested |