Benchmark Almanac

Earth-Bench-Pro

Earth-Agent-Pro + GPT-5 66.1% tops its leaderboard
Not in model cards

Agents complete Earth observation workflows under three conditions. They follow instructions, plan from supplied observations, or obtain observations at runtime.

The authors derive 744 questions from 248 shared tasks. They report final answer quality and agreement with the reference tool calls.

Released
September 2026
Kind
Apps and MCP servers
Level
University
Scored
Graded by a model
Answers
Agent task
Tasks
All public
Measures
Capability
Language
English
Interaction
Agentic
Environment
APIs and tools
Refresh
One fixed set
Run by
Anyone
Based on
Earth-Bench
In views
Agents, Vision
21
Entries in its paper, 0 models we track

Earth-Agent-Pro + GPT-5 tops the leaderboard with 66.1% No scores yet

Best score per entry from the benchmark's paper, read on November 20th, 2026. A paper’s table doesn’t add newer models. The headline is judged answer accuracy on the 248 Open-World Execution questions. Where a model ran on its own with several settings, we keep its best. Agents other companies built around a model get a pale shade of that model's lab color. Gray marks blends and entries not in our model list, such as fine-tunes and older models. None of these count toward leaders. Agents built around a model, such as Earth-Agent-Pro + GPT-5, are their own entries. We don't count them as the model's score. This view shows only models we track, each at its best.

  1. 1 Earth-Agent-Pro + GPT-5 66.1%
  2. 2 Earth-Agent-Pro + Gemini 3.5 Flash 63.3%
  3. 3 Earth-Agent-Pro + Claude Sonnet 4.6 56.9%
  4. 4 Earth-Agent-Pro + Qwen3.5-9B Fine-tuned planner and executor 50%
  5. 5 Earth-Agent-Pro + Qwen3.5-27B 49.6%
  6. 6 Earth-Agent-Pro + GLM-4.5-Air 48.4%
  7. 7 Earth-Agent-Pro + gpt-oss-120b 48.4%
  8. 8 Earth-Agent-Pro + Gemma 4 26B A4B 47.6%
  9. 9 ReAct + GPT-5 45.2%
  10. 10 Earth-Agent-Pro + Qwen3.5-4B Fine-tuned planner and executor 43.1%
  11. 11 OpenEarthAgent + GPT-5 41.5%
  12. 12 Earth-Agent-Pro + Intern-S1 40.7%
  13. 13 AFlow + GPT-5 40.2%
  14. 14 Earth-Agent-Pro + gpt-oss-20b 38.7%
  15. 15 Earth-Agent-Pro + Llama 3.3 70B Instruct 33.9%
  16. 16 Earth-Agent-Pro + Llama 4 Scout 32.7%
  17. 17 Earth-Agent-Pro + Gemma 4 E4B 31.4%
  18. 18 Earth-Agent-Pro + InternLM3-8B-Instruct 19.4%
  19. 19 MistralAI-v0.3-7B 14.1%
  20. 20 Llama-3.1-8B-Instruct 13.7%
  21. 21 Llama-3.2-3B-Instruct 4.44%

Source: The Earth-Bench-Pro paper

#EntryScoreSettingDateSource
1 Earth-Agent-Pro + GPT-5 66.1% Aug 2025 Official leaderboard
2 Earth-Agent-Pro + Gemini 3.5 Flash 63.3% May 2026 Official leaderboard
3 Earth-Agent-Pro + Claude Sonnet 4.6 56.9% Feb 2026 Official leaderboard
4 Earth-Agent-Pro + Qwen3.5-9B 50% Fine-tuned planner and executor Feb 2026 Official leaderboard
5 Earth-Agent-Pro + Qwen3.5-27B 49.6% Feb 2026 Official leaderboard
6 Earth-Agent-Pro + GLM-4.5-Air 48.4% Jul 2025 Official leaderboard
7 Earth-Agent-Pro + gpt-oss-120b 48.4% Aug 2025 Official leaderboard
8 Earth-Agent-Pro + Gemma 4 26B A4B 47.6% Apr 2026 Official leaderboard
9 ReAct + GPT-5 45.2% Aug 2025 Official leaderboard
10 Earth-Agent-Pro + Qwen3.5-4B 43.1% Fine-tuned planner and executor Feb 2026 Official leaderboard
11 OpenEarthAgent + GPT-5 41.5% Aug 2025 Official leaderboard
12 Earth-Agent-Pro + Intern-S1 40.7% Jul 2025 Official leaderboard
13 AFlow + GPT-5 40.2% Aug 2025 Official leaderboard
14 Earth-Agent-Pro + gpt-oss-20b 38.7% Aug 2025 Official leaderboard
15 Earth-Agent-Pro + Llama 3.3 70B Instruct 33.9% Dec 2024 Official leaderboard
16 Earth-Agent-Pro + Llama 4 Scout 32.7% Apr 2025 Official leaderboard
17 Earth-Agent-Pro + Gemma 4 E4B 31.4% Mar 2026 Official leaderboard
18 Earth-Agent-Pro + InternLM3-8B-Instruct 19.4% Jan 2025 Official leaderboard
19 MistralAI-v0.3-7B 14.1% Earth-Agent-Pro Official leaderboard
20 Llama-3.1-8B-Instruct 13.7% Earth-Agent-Pro Official leaderboard
21 Llama-3.2-3B-Instruct 4.44% Earth-Agent-Pro Official leaderboard
#ModelScoreSettingDateSource

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.

More in Tool use

All 265 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 8 checks pass. Every benchmark

Catalog record 2 pass
  • Pass Task access All tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 2 pass 3 notes
  • Pass Independent runs Scores from the benchmark’s paper (21 entries).
  • Pass Models scored 21 entries on its leaderboard, none a model we track.
  • Note Newest score No scores yet.
  • Note Frontier models No models scored yet.
  • Note Headroom No scores yet.
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.

How we check