Earth-Bench
Version 1
Agents analyze satellite observations with Earth science tools. The evaluator checks the final answer and compares tool calls with a reference sequence.
The authors test 248 tasks with prepared inputs. Agents either follow prescribed steps or plan their own workflow.
- Made by
- OpenDataLab and collaborators
- Released
- September 2025
- Kind
- Apps and MCP servers
- Level
- University
- Scored
- Checked automatically
- Answers
- Agent task
- Tasks
- All public
- Measures
- Capability
- Subject
- Earth and environmental sciences
- Language
- English
- In and out
- Text, Images, Maps and geodata → Actions, Text
- Interaction
- Agentic
- Environment
- APIs and tools
- Items from
- Real work and use
- Refresh
- New versions replace it
- Run by
- Anyone
- Strongest model tested
- GPT-5 #71 on the capability index
- Last model added
- September 2025
- Scores refreshed
- November 20th, 2026
GPT-5 leads with 66% among models released by September 2025 GPT-5 leads with 66% among models released by September 2025
Best score per entry from the benchmark's paper, read on November 20th, 2026. A paper’s table doesn’t add newer models. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders. This view shows only models we track, each at its best.
The newest model with a score here came out in September 2025, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- GPT-6.2 Astra
- Claude Fable 5.2
-
Gemini 4 Carbon
- GPT-6 Astra
- and 16 more
- 1 GPT-5 Auto-planning 66%
- 2
Doubao-Seed-1.6 Instruction-following 59.5%
- 3 Kimi K2 Sep 2025, Instruction-following 56.7%
- 4
Gemini 2.5 Flash Sep 2025, Instruction-following 53%
- 5
DeepSeek-V3.1 Instruction-following 52.2%
- 6
Qwen3-Max-Preview Auto-planning 50.2%
- 7 GPT-4o (snapshot unspecified) Instruction-following 44.9%
- 8
Llama 4 Maverick Auto-planning 44.9%
- 9
Qwen Plus Apr 2025, Auto-planning 42.5%
- 10 GLM-4.5V Instruction-following 35.3%
- 11 Mistral-Large (snapshot unspecified) Auto-planning 30%
- 12 InternVL-3.5 Auto-planning 26.7%
- 13
Qwen3-32B Instruction-following 24.8%
Source: The Earth-Bench paper
- 1 GPT-5 Auto-planning 66%
- 2
Doubao-Seed-1.6 Instruction-following 59.5%
- 3 Kimi K2 Sep 2025, Instruction-following 56.7%
- 4
Gemini 2.5 Flash Sep 2025, Instruction-following 53%
- 5
DeepSeek-V3.1 Instruction-following 52.2%
- 6
Qwen3-Max-Preview Auto-planning 50.2%
- 7
Llama 4 Maverick Auto-planning 44.9%
- 8
Qwen Plus Apr 2025, Auto-planning 42.5%
- 9 GLM-4.5V Instruction-following 35.3%
- 10
Qwen3-32B Instruction-following 24.8%
Source: The Earth-Bench paper
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | GPT-5 | 66% | Auto-planning, Earth-Agent | Aug 2025 | Official leaderboard |
| 2 | | 59.5% | Instruction-following, Earth-Agent | Jun 2025 | Official leaderboard |
| 3 | Kimi K2 Sep 2025 | 56.7% | Instruction-following, Earth-Agent | Sep 2025 | Official leaderboard |
| 4 | | 53% | Instruction-following, Earth-Agent | Sep 2025 | Official leaderboard |
| 5 | | 52.2% | Instruction-following, Earth-Agent | Aug 2025 | Official leaderboard |
| 6 | | 50.2% | Auto-planning, Earth-Agent | Sep 2025 | Official leaderboard |
| 7 | GPT-4o (snapshot unspecified) | 44.9% | Instruction-following, Earth-Agent | Official leaderboard | |
| 8 | | 44.9% | Auto-planning, Earth-Agent | Apr 2025 | Official leaderboard |
| 9 | | 42.5% | Auto-planning, Earth-Agent | Apr 2025 | Official leaderboard |
| 10 | GLM-4.5V | 35.3% | Instruction-following, Earth-Agent | Aug 2025 | Official leaderboard |
| 11 | Mistral-Large (snapshot unspecified) | 30% | Auto-planning, Earth-Agent | Official leaderboard | |
| 12 | InternVL-3.5 | 26.7% | Auto-planning, Earth-Agent | Official leaderboard | |
| 13 | | 24.8% | Instruction-following, Earth-Agent | Apr 2025 | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | GPT-5 | 66% | Auto-planning, Earth-Agent | Aug 2025 | Official leaderboard |
| 2 | | 59.5% | Instruction-following, Earth-Agent | Jun 2025 | Official leaderboard |
| 3 | Kimi K2 Sep 2025 | 56.7% | Instruction-following, Earth-Agent | Sep 2025 | Official leaderboard |
| 4 | | 53% | Instruction-following, Earth-Agent | Sep 2025 | Official leaderboard |
| 5 | | 52.2% | Instruction-following, Earth-Agent | Aug 2025 | Official leaderboard |
| 6 | | 50.2% | Auto-planning, Earth-Agent | Sep 2025 | Official leaderboard |
| 7 | | 44.9% | Auto-planning, Earth-Agent | Apr 2025 | Official leaderboard |
| 8 | | 42.5% | Auto-planning, Earth-Agent | Apr 2025 | Official leaderboard |
| 9 | GLM-4.5V | 35.3% | Instruction-following, Earth-Agent | Aug 2025 | Official leaderboard |
| 10 | | 24.8% | Instruction-following, Earth-Agent | Apr 2025 | Official leaderboard |
Rankings by part
We read 3 parts from the benchmark’s paper: Spectrum, Products and RGB. Each ranks models on one part of the test. GPT-5 leads 2 of them, and Kimi K2 leads the other one. Arrows show how far a model moved from its overall rank among the models we track.
Spectrum
10 models we track have a score here.
- 1 GPT-5 64%
- 2
Doubao-Seed-1.6 57%
- 3
Gemini 2.5 Flash Sep 2025 ↑1 52%
- 4 Kimi K2 Sep 2025 ↓1 50%
- 5
DeepSeek-V3.1 47%
- 6
Qwen3-Max-Preview 45%
- 7
Llama 4 Maverick 37%
- 8
Qwen Plus Apr 2025 36%
- 9 GLM-4.5V 33.3%
- 10
Qwen3-32B 29%
Products
10 models we track have a score here.
- 1 GPT-5 75%
- 2
Doubao-Seed-1.6 67%
- 3
Gemini 2.5 Flash Sep 2025 ↑1 63.6%
- 4 Kimi K2 Sep 2025 ↓1 62.5%
- 5
Qwen3-Max-Preview ↑1 61.4%
- 6
Llama 4 Maverick ↑1 60.2%
- 7
DeepSeek-V3.1 ↓2 59.1%
- 8
Qwen Plus Apr 2025 53.4%
- 9 GLM-4.5V 47.7%
- 10
Qwen3-32B 27.3%
RGB
10 models we track have a score here.
- 1 Kimi K2 Sep 2025 ↑2 62.7%
- 2 GPT-5 ↓1 59.3%
- 3
Doubao-Seed-1.6 ↓1 55.9%
- 4
DeepSeek-V3.1 ↑1 55.9%
- 5
Qwen3-Max-Preview ↑1 49.1%
- 6
Gemini 2.5 Flash Sep 2025 ↓2 47.5%
- 7
Qwen Plus Apr 2025 ↑1 42.4%
- 8
Llama 4 Maverick ↓1 37.3%
- 9 GLM-4.5V 28.8%
- 10
Qwen3-32B 27.1%
Scores at each effort level
Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.
Source: The Earth-Bench paper
| Model | Effort | Harness | Score | Cost per task | Output tokens |
|---|---|---|---|---|---|
| GPT-5 | Auto-planning | Earth-Agent | 66% | ||
| GPT-5 | Instruction-following | Earth-Agent | 62.4% | ||
| Instruction-following | Earth-Agent | 59.5% | |||
| Kimi K2 Sep 2025 | Instruction-following | Earth-Agent | 56.7% | ||
| Instruction-following | Earth-Agent | 53% | |||
| Auto-planning | Earth-Agent | 52.5% | |||
| Instruction-following | Earth-Agent | 52.2% | |||
| Auto-planning | Earth-Agent | 52.2% | |||
| Auto-planning | Earth-Agent | 51.4% | |||
| Kimi K2 Sep 2025 | Auto-planning | Earth-Agent | 50.6% | ||
| Auto-planning | Earth-Agent | 50.2% | |||
| Instruction-following | Earth-Agent | 47.4% | |||
| Auto-planning | Earth-Agent | 44.9% | |||
| Auto-planning | Earth-Agent | 42.5% | |||
| Instruction-following | Earth-Agent | 38.5% | |||
| Instruction-following | Earth-Agent | 38.5% | |||
| GLM-4.5V | Instruction-following | Earth-Agent | 35.3% | ||
| GLM-4.5V | Auto-planning | Earth-Agent | 32.9% | ||
| Instruction-following | Earth-Agent | 24.8% | |||
| Auto-planning | Earth-Agent | 20.6% |
Scores at each effort level
Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.
Source: The Earth-Bench paper
| Model | Effort | Harness | Score | Cost per task | Output tokens |
|---|---|---|---|---|---|
| GPT-5 | Auto-planning | Earth-Agent | 66% | ||
| GPT-5 | Instruction-following | Earth-Agent | 62.4% | ||
| Instruction-following | Earth-Agent | 59.5% | |||
| Kimi K2 Sep 2025 | Instruction-following | Earth-Agent | 56.7% | ||
| Instruction-following | Earth-Agent | 53% | |||
| Auto-planning | Earth-Agent | 52.5% | |||
| Instruction-following | Earth-Agent | 52.2% | |||
| Auto-planning | Earth-Agent | 52.2% | |||
| Auto-planning | Earth-Agent | 51.4% | |||
| Kimi K2 Sep 2025 | Auto-planning | Earth-Agent | 50.6% | ||
| Auto-planning | Earth-Agent | 50.2% | |||
| Instruction-following | Earth-Agent | 47.4% | |||
| Auto-planning | Earth-Agent | 44.9% | |||
| Auto-planning | Earth-Agent | 42.5% | |||
| Instruction-following | Earth-Agent | 38.5% | |||
| Instruction-following | Earth-Agent | 38.5% | |||
| GLM-4.5V | Instruction-following | Earth-Agent | 35.3% | ||
| GLM-4.5V | Auto-planning | Earth-Agent | 32.9% | ||
| Instruction-following | Earth-Agent | 24.8% | |||
| Auto-planning | Earth-Agent | 20.6% |
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: The Earth-Bench paper
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official Earth-Bench leaderboard
No lab model card reports it yet
We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.
Versions
We switch the top row to each new version when it comes out. Open it for one row per version. An old version stays in play while people still test it, so whoever first takes it past 85% gets credit for closing it out.
Source: Official leaderboards
More in Tool use
- AutomationBench In use Agent completes end-to-end business workflows inside a simulated company exposing dozens of REST APIs. Graded on resulting state.
Gemini 4 Neon 56.3%
- MCP Atlas In use Multi-step workflows over real MCP servers. Scores tool-use competency on realistic tasks.
Gemini 4 Argon 90.7%
- BFCL V4 In use Models call functions and tools, including multi-turn agentic tasks with web search and memory. Checked against expected calls and outcomes.
Qwen3.7-Max 75%
- τ³-bench Banking In use Agents act as banking customer-service reps over many turns with a simulated user, following policy. Pass^1 on the final database state.
Qwen 3.8 Max 55.1%
- Toolathlon-Verified In use 108 long-horizon tasks across 32 apps through about 600 MCP tools. Execution-based checkers inspect final environment state. GLM-5.3-Flash 78.4%
- DeepSearchQA In use Questions that need multi-step search and synthesis. Final answers graded against references. Kimi K3 95%
Data audit
All 9 checks pass. Every benchmark
Catalog record 2 pass
- Pass Task access All tasks are public.
- Pass Confidence Checked against a primary source.
Scores 3 pass 2 notes
- Pass Independent runs Scores from the benchmark’s paper (13 entries).
- Pass Models scored 10 models we track, plus 3 other entries on its leaderboard.
- Note Newest score The newest came 14 months ago, in September 2025, for Gemini 2.5 Flash. That’s expected for a test replaced by Earth-Bench V2.
- Note Frontier models None of the top 10 has a score, which is fine for a retired test. The strongest tested is GPT-5 (#71).
- Pass Headroom The best is 66%, under 85%, so it still separates models.
Leaderboard import 3 pass
- Pass Importer We read the board on arxiv.org: 20 rows for models we track and 3 other entries at the last read.
- Pass Last read November 20th, 2026.
- Pass Read worked The last read worked.
Consistency 1 pass 1 note
- Note Lab claims No model has both a lab number and an independent run.
- Pass Known issues No published audit found problems.