TerraBench: leaderboard
TerraBench (2026) evaluates tool-using agents on 403 executable Earth-science tasks that combine Earth-observation imagery, gridded climate data, GIS operations and simulators, each model driving the TerraAgent ReAct harness with 77 scientific tools. The score is Hit@tol, the percent of answer fields within their tolerance band, with exact match for strings and booleans. Higher is better.
Metric: Hit@tol (%): share of answer fields within the field's tolerance band (exact match for strings and booleans), averaged over multi-field outputs, on the 403 executable TerraBench tasks (Fundamentals, Simulator-Grounded and Document-Grounded Verification tracks across eight Earth-science domains), each model driving the benchmark's TerraAgent ReAct harness with its 77 scientific tools; higher is better. Source: arxiv.org. Saturation forecast: Around November 2027. 17 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 4.6 | 22.88 | #161 |
| 2 | GPT-5.5 | 21.03 | #49 |
| 3 | Claude Haiku 4.5 | 15.23 | #568 |
| 4 | Gemini 3.1 Pro (Preview) | 13.21 | #94 |
| 5 | GPT-5.4 | 9.8 | #123 |
| 6 | Gemini 2.5 Flash | 6.34 | #453 |
| 7 | Qwen 3.5 35B A3B | 5.89 | #482 |
| 8 | Qwen 3 14B | 3.8 | #1233 |
| 9 | Gemma 4 26B A4B | 3.67 | #615 |
| 10 | Gemma 4 E4B | 2.9 | #2214 |
| 11 | Mistral 7B Instruct (v0.3) | 1.67 | #6653 |
| 12 | Qwen 3 8B | 1.33 | #1830 |
| 13 | Qwen 3.5 9B | 1.2 | #803 |
| 14 | InternVL3-8B | 1.1 | #1688 |
| 15 | Qwen 3 1.7B | 1.08 | #5520 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=terrabench · How It Works · Data refreshed daily, snapshot 2026-11-13.