TerraLogic: leaderboard
TerraLogic (CUHK-Shenzhen and TUM, 2026) tests hierarchical geospatial reasoning on 545 scenario-driven tasks over optical, SAR and infrared remote-sensing imagery, solved through verifiable perception and spatial-analysis tool chains with each model driving the HieraPlan reference agent. The score is final-answer accuracy in percent, judged by GPT-4o-mini against the reference; higher is better.
Metric: Final answer accuracy (%, AnsAcc) over the 545 TerraLogic optical, SAR and infrared remote-sensing tasks, judged by GPT-4o-mini with a rubric against the ground-truth answer and its key arguments; each model drives the HieraPlan tool agent under a ReAct-style protocol in OpenCompass; higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 12 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4o | 25.7 | #741 |
| 2 | GPT-3.5 | 24.4 | #2417 |
| 3 | Llama 3 70B Instruct | 22.75 | #1676 |
| 4 | Qwen 2.5 7B Instruct | 22.57 | #2564 |
| 5 | DeepSeek V3 | 21.28 | #716 |
| 6 | Qwen 2.5 32B Instruct | 19.08 | #1215 |
| 7 | InternLM3-8B-Instruct | 18.9 | #2568 |
| 8 | Mistral 7B Instruct (v0.2) | 13.21 | #7007 |
| 9 | Gemini 2.5 Flash | 12.04 | #453 |
| 10 | Yi-1.5-6B-Chat | 10.83 | #6399 |
| 11 | Phi-3 Mini 4K Instruct | 3.67 | #6810 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=terralogic · How It Works · Data refreshed daily, snapshot 2026-11-13.