TerraBench - NumScore: leaderboard
Metric: NumScore (%): tolerance-aware partial credit for numeric answers, halving for each extra tolerance width of error (missing or unparseable answers score 0), 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 April 2028. 17 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Sonnet 4.6 | 28.44 |
| 2 | GPT-5.5 | 26.45 |
| 3 | Claude Haiku 4.5 | 21.19 |
| 4 | Gemini 3.1 Pro (Preview) | 17.07 |
| 5 | GPT-5.4 | 12.29 |
| 6 | Gemini 2.5 Flash | 7.7 |
| 7 | Qwen 3.5 35B A3B | 7.49 |
| 8 | Gemma 4 26B A4B | 4.45 |
| 9 | Qwen 3 14B | 4.3 |
| 10 | Gemma 4 E4B | 3.04 |
| 11 | Mistral 7B Instruct (v0.3) | 2.41 |
| 12 | Qwen 3 8B | 2.06 |
| 13 | InternVL3-8B | 1.7 |
| 14 | Qwen 3.5 9B | 1.3 |
| 15 | Qwen 3 1.7B | 1.15 |
Interactive version: theaggregate.ai/benchmark?slug=terrabench-numscore · How It Works · Data refreshed daily, snapshot 2026-09-29.