Earth Observation Agent Task Completion on Earth-Bench Lite
65Spectrum ScoreEarth-Agent
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Earth-AgentBackbone=GPT5, Latency=158 min2025.09 | 65 | 36.84 | 65.71 | 55.83 | |
| Earth-AgentBackbone=Deepseek-V3.1, Latency=79 min2025.09 | 50 | 42.11 | 51.43 | 47.84 | |
| GPT-AgentLatency=≈ 300 min2025.09 | 45 | 31.6 | 45.26 | 40.42 | |
| MGXLatency=≈ 60 min2025.09 | 40 | 15.8 | 0 | 18.6 | |
| Earth-AgentBackbone=Kimik2, Latency=131 min2025.09 | 36.84 | 50 | 50 | 45.95 | |
| CozeLatency=≈ 120 min2025.09 | 35 | 10.5 | 0 | 15.3 | |
| ManusLatency=≈ 150 min2025.09 | 15 | 15.8 | 47.62 | 26.14 |