AgriEval (Harbin Institute of Technology)
Models answer Chinese agricultural exam questions across 29 subfields. The authors score 14,697 choice questions by accuracy and 2,167 open answers by ROUGE-L.
AgriJev independently evaluated choice questions with Qwen models. The two released Chinese files match the paper counts. The repository overview gives different totals.
- Released
- May 2025
- Kind
- Exam questions
- Level
- University
- Scored
- Checked automatically
- Answers
- Multiple choice
- Tasks
- All public
- Measures
- Capability
- Language
- Chinese
- Interaction
- Single turn
- Items from
- Exams and contests
- Refresh
- One fixed set
- Run by
- Anyone
- Strongest model tested
-
Gemini 2.0 Flash Feb 2025 #147 on the capability index
- Last model added
- April 2025
- Scores refreshed
- November 20th, 2026
Qwen Plus leads with 63.2% among models released by April 2025 Qwen Plus leads with 63.2% among models released by April 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 April 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
- 63.2
Qwen Plus
- 62.7
Qwen2-72B
- 60.3
Qwen2.5-72B
- 59.3 DeepSeek-V3 (snapshot unspecified)
- 56.8 Qwen2.5-32B-Instruct
- 56.3
Qwen-Turbo
- 55.4 Yi-1.5-34b-Chat
- 55.3
Gemini 2.0 Flash
- 54.9 Claude-3.5-Sonnet (snapshot unspecified)
- 53.0
Qwen1.5-72B
- 52.6 GPT-4o (snapshot unspecified)
- 52.1 Qwen2.5-14B-Instruct
- 50.0 GLM-4-Air
- 49.6 GPT-4o-mini (snapshot unspecified)
- 49.2 Qwen2.5-7B-Instruct
- 48.8
Qwen2-7B-Instruct
- 48.4 Qwen-72B-Chat
- 47.8 Qwen1.5-32B-Chat
- 45.5 GLM-4-Flash
- 45.1
Qwen1.5-14B-Chat
- 43.8 GLM-4-9B-Chat
- 43.8 InternLM2.5-7B-Chat
- 43.2 Llama3-70B-Instruct
- 42.3
InternLM2.5-20B-Chat
- 41.3
Yi-1.5-9B-Chat
- 40.7 Qwen2.5-3B-Instruct
- 40.5
Qwen-14B
- 39.5 Marco-o1
- 39.4 Qwen2.5-Coder-7B-Instruct
- 38.4
Qwen1.5-7B-Chat
- 36.9
phi-3-medium 14B
- 36.8 GPT-3.5-Turbo (snapshot unspecified)
- 35.5 ChatGLM3-6B
- 35.5 Baichuan2-13B-Chat
- 35.0 Qwen-7B-Chat
- 34.4 Mistral-Nemo-Instruct
- 34.2 DeepSeek-LLM-7B-Chat
- 34.2
Phi-3.5-mini
- 34.2
Qwen1.5-4B-Chat
- 33.6 InternLM2-Chat-7B
- 32.8 chatglm2-6b
- 32.5 Baichuan2-7B-Chat
- 32.3 Qwen1.5-1.8B-Chat
- 31.4 Llama3-8B-Instruct
- 30.4
Mistral-7B-Instruct-v0.1
- 27.1 Llama2-70B-Chat
- 26.6
Baichuan2-13B
- 25.4
Baichuan 2-7B
- 23.9 Llama2-7B-Chat
- 21.1 Llama2-13B-Chat
- 20.5 DeepSeek-LLM-7B-Base
- 1
Qwen Plus Apr 2025, Zero-shot 63.2%
- 2
Qwen2-72B Zero-shot 62.7%
- 3
Qwen2.5-72B Zero-shot 60.3%
- 4 DeepSeek-V3 (snapshot unspecified) Few-shot 59.3%
- 5 Qwen2.5-32B-Instruct Few-shot 56.8%
- 6
Qwen-Turbo Few-shot 56.3%
- 7 Yi-1.5-34b-Chat Few-shot 55.4%
- 8
Gemini 2.0 Flash Feb 2025, Zero-shot 55.3%
- 9 Claude-3.5-Sonnet (snapshot unspecified) Zero-shot 54.9%
- 10
Qwen1.5-72B Zero-shot 53%
- 11 GPT-4o (snapshot unspecified) Few-shot 52.6%
- 12 Qwen2.5-14B-Instruct Few-shot 52.1%
- 13 GLM-4-Air Zero-shot 50%
- 14 GPT-4o-mini (snapshot unspecified) Few-shot 49.6%
- 15 Qwen2.5-7B-Instruct Few-shot 49.2%
- 16
Qwen2-7B-Instruct Zero-shot 48.8%
- 17 Qwen-72B-Chat Chain of thought 48.4%
- 18 Qwen1.5-32B-Chat Few-shot 47.8%
- 19 GLM-4-Flash Zero-shot 45.5%
- 20
Qwen1.5-14B-Chat Zero-shot 45.1%
- 21 GLM-4-9B-Chat Chain of thought 43.8%
- 22 InternLM2.5-7B-Chat Zero-shot 43.8%
- 23 Llama3-70B-Instruct Zero-shot 43.2%
- 24
InternLM2.5-20B-Chat Zero-shot 42.3%
- 25
Yi-1.5-9B-Chat Chain of thought 41.3%
- 26 Qwen2.5-3B-Instruct Few-shot 40.7%
- 27
Qwen-14B Chain of thought 40.5%
- 28 Marco-o1 Few-shot 39.5%
- 29 Qwen2.5-Coder-7B-Instruct Chain of thought 39.4%
- 30
Qwen1.5-7B-Chat Chain of thought 38.4%
- 31
phi-3-medium 14B Zero-shot 36.9%
- 32 GPT-3.5-Turbo (snapshot unspecified) Chain of thought 36.8%
- 33 ChatGLM3-6B Zero-shot 35.5%
- 34 Baichuan2-13B-Chat Zero-shot 35.5%
- 35 Qwen-7B-Chat Zero-shot 35%
- 36 Mistral-Nemo-Instruct Zero-shot 34.4%
- 37 DeepSeek-LLM-7B-Chat Chain of thought 34.2%
- 38
Phi-3.5-mini Zero-shot 34.2%
- 39
Qwen1.5-4B-Chat Zero-shot 34.2%
- 40 InternLM2-Chat-7B Zero-shot 33.6%
- 41 chatglm2-6b Zero-shot 32.8%
- 42 Baichuan2-7B-Chat Zero-shot 32.5%
- 43 Qwen1.5-1.8B-Chat Zero-shot 32.3%
- 44 Llama3-8B-Instruct Zero-shot 31.4%
- 45
Mistral-7B-Instruct-v0.1 Few-shot 30.4%
- 46 Llama2-70B-Chat Zero-shot 27.1%
- 47
Baichuan2-13B Chain of thought 26.6%
- 48
Baichuan 2-7B Chain of thought 25.4%
- 49 Llama2-7B-Chat Zero-shot 23.9%
- 50 Llama2-13B-Chat Zero-shot 21.1%
- 51 DeepSeek-LLM-7B-Base Chain of thought 20.5%
Source: The AgriEval (Harbin Institute of Technology) paper
- 63.2
Qwen Plus
- 62.7
Qwen2-72B
- 60.3
Qwen2.5-72B
- 56.3
Qwen-Turbo
- 55.3
Gemini 2.0 Flash
- 53.0
Qwen1.5-72B
- 48.8
Qwen2-7B-Instruct
- 45.1
Qwen1.5-14B-Chat
- 43.8 GLM-4-9B-Chat
- 42.3
InternLM2.5-20B-Chat
- 41.3
Yi-1.5-9B-Chat
- 40.5
Qwen-14B
- 38.4
Qwen1.5-7B-Chat
- 36.9
phi-3-medium 14B
- 35.5 ChatGLM3-6B
- 35.5 Baichuan2-13B-Chat
- 34.2
Phi-3.5-mini
- 34.2
Qwen1.5-4B-Chat
- 32.8 chatglm2-6b
- 30.4
Mistral-7B-Instruct-v0.1
- 26.6
Baichuan2-13B
- 25.4
Baichuan 2-7B
- 1
Qwen Plus Apr 2025, Zero-shot 63.2%
- 2
Qwen2-72B Zero-shot 62.7%
- 3
Qwen2.5-72B Zero-shot 60.3%
- 4
Qwen-Turbo Few-shot 56.3%
- 5
Gemini 2.0 Flash Feb 2025, Zero-shot 55.3%
- 6
Qwen1.5-72B Zero-shot 53%
- 7
Qwen2-7B-Instruct Zero-shot 48.8%
- 8
Qwen1.5-14B-Chat Zero-shot 45.1%
- 9 GLM-4-9B-Chat Chain of thought 43.8%
- 10
InternLM2.5-20B-Chat Zero-shot 42.3%
- 11
Yi-1.5-9B-Chat Chain of thought 41.3%
- 12
Qwen-14B Chain of thought 40.5%
- 13
Qwen1.5-7B-Chat Chain of thought 38.4%
- 14
phi-3-medium 14B Zero-shot 36.9%
- 15 ChatGLM3-6B Zero-shot 35.5%
- 16 Baichuan2-13B-Chat Zero-shot 35.5%
- 17
Phi-3.5-mini Zero-shot 34.2%
- 18
Qwen1.5-4B-Chat Zero-shot 34.2%
- 19 chatglm2-6b Zero-shot 32.8%
- 20
Mistral-7B-Instruct-v0.1 Few-shot 30.4%
- 21
Baichuan2-13B Chain of thought 26.6%
- 22
Baichuan 2-7B Chain of thought 25.4%
Source: The AgriEval (Harbin Institute of Technology) paper
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | | 63.2% | Zero-shot | Apr 2025 | Official leaderboard |
| 2 | | 62.7% | Zero-shot | Jun 2024 | Official leaderboard |
| 3 | | 60.3% | Zero-shot | Sep 2024 | Official leaderboard |
| 4 | DeepSeek-V3 (snapshot unspecified) | 59.3% | Few-shot | Official leaderboard | |
| 5 | Qwen2.5-32B-Instruct | 56.8% | Few-shot | Official leaderboard | |
| 6 | | 56.3% | Few-shot | Nov 2024 | Official leaderboard |
| 7 | Yi-1.5-34b-Chat | 55.4% | Few-shot | Official leaderboard | |
| 8 | | 55.3% | Zero-shot | Feb 2025 | Official leaderboard |
| 9 | Claude-3.5-Sonnet (snapshot unspecified) | 54.9% | Zero-shot | Official leaderboard | |
| 10 | | 53% | Zero-shot | Feb 2024 | Official leaderboard |
| 11 | GPT-4o (snapshot unspecified) | 52.6% | Few-shot | Official leaderboard | |
| 12 | Qwen2.5-14B-Instruct | 52.1% | Few-shot | Official leaderboard | |
| 13 | GLM-4-Air | 50% | Zero-shot | Official leaderboard | |
| 14 | GPT-4o-mini (snapshot unspecified) | 49.6% | Few-shot | Official leaderboard | |
| 15 | Qwen2.5-7B-Instruct | 49.2% | Few-shot | Official leaderboard | |
| 16 | | 48.8% | Zero-shot | Jun 2024 | Official leaderboard |
| 17 | Qwen-72B-Chat | 48.4% | Chain of thought | Official leaderboard | |
| 18 | Qwen1.5-32B-Chat | 47.8% | Few-shot | Official leaderboard | |
| 19 | GLM-4-Flash | 45.5% | Zero-shot | Official leaderboard | |
| 20 | | 45.1% | Zero-shot | Feb 2024 | Official leaderboard |
| 21 | GLM-4-9B-Chat | 43.8% | Chain of thought | Jun 2024 | Official leaderboard |
| 22 | InternLM2.5-7B-Chat | 43.8% | Zero-shot | Official leaderboard | |
| 23 | Llama3-70B-Instruct | 43.2% | Zero-shot | Official leaderboard | |
| 24 | | 42.3% | Zero-shot | Aug 2024 | Official leaderboard |
| 25 | | 41.3% | Chain of thought | May 2024 | Official leaderboard |
| 26 | Qwen2.5-3B-Instruct | 40.7% | Few-shot | Official leaderboard | |
| 27 | | 40.5% | Chain of thought | Sep 2023 | Official leaderboard |
| 28 | Marco-o1 | 39.5% | Few-shot | Official leaderboard | |
| 29 | Qwen2.5-Coder-7B-Instruct | 39.4% | Chain of thought | Official leaderboard | |
| 30 | | 38.4% | Chain of thought | Feb 2024 | Official leaderboard |
| 31 | | 36.9% | Zero-shot | Apr 2024 | Official leaderboard |
| 32 | GPT-3.5-Turbo (snapshot unspecified) | 36.8% | Chain of thought | Official leaderboard | |
| 33 | ChatGLM3-6B | 35.5% | Zero-shot | Oct 2023 | Official leaderboard |
| 34 | Baichuan2-13B-Chat | 35.5% | Zero-shot | Sep 2023 | Official leaderboard |
| 35 | Qwen-7B-Chat | 35% | Zero-shot | Official leaderboard | |
| 36 | Mistral-Nemo-Instruct | 34.4% | Zero-shot | Official leaderboard | |
| 37 | DeepSeek-LLM-7B-Chat | 34.2% | Chain of thought | Official leaderboard | |
| 38 | | 34.2% | Zero-shot | Aug 2024 | Official leaderboard |
| 39 | | 34.2% | Zero-shot | Feb 2024 | Official leaderboard |
| 40 | InternLM2-Chat-7B | 33.6% | Zero-shot | Official leaderboard | |
| 41 | chatglm2-6b | 32.8% | Zero-shot | Jun 2023 | Official leaderboard |
| 42 | Baichuan2-7B-Chat | 32.5% | Zero-shot | Official leaderboard | |
| 43 | Qwen1.5-1.8B-Chat | 32.3% | Zero-shot | Official leaderboard | |
| 44 | Llama3-8B-Instruct | 31.4% | Zero-shot | Official leaderboard | |
| 45 | | 30.4% | Few-shot | Sep 2023 | Official leaderboard |
| 46 | Llama2-70B-Chat | 27.1% | Zero-shot | Official leaderboard | |
| 47 | | 26.6% | Chain of thought | Sep 2023 | Official leaderboard |
| 48 | | 25.4% | Chain of thought | Sep 2023 | Official leaderboard |
| 49 | Llama2-7B-Chat | 23.9% | Zero-shot | Official leaderboard | |
| 50 | Llama2-13B-Chat | 21.1% | Zero-shot | Official leaderboard | |
| 51 | DeepSeek-LLM-7B-Base | 20.5% | Chain of thought | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | | 63.2% | Zero-shot | Apr 2025 | Official leaderboard |
| 2 | | 62.7% | Zero-shot | Jun 2024 | Official leaderboard |
| 3 | | 60.3% | Zero-shot | Sep 2024 | Official leaderboard |
| 4 | | 56.3% | Few-shot | Nov 2024 | Official leaderboard |
| 5 | | 55.3% | Zero-shot | Feb 2025 | Official leaderboard |
| 6 | | 53% | Zero-shot | Feb 2024 | Official leaderboard |
| 7 | | 48.8% | Zero-shot | Jun 2024 | Official leaderboard |
| 8 | | 45.1% | Zero-shot | Feb 2024 | Official leaderboard |
| 9 | GLM-4-9B-Chat | 43.8% | Chain of thought | Jun 2024 | Official leaderboard |
| 10 | | 42.3% | Zero-shot | Aug 2024 | Official leaderboard |
| 11 | | 41.3% | Chain of thought | May 2024 | Official leaderboard |
| 12 | | 40.5% | Chain of thought | Sep 2023 | Official leaderboard |
| 13 | | 38.4% | Chain of thought | Feb 2024 | Official leaderboard |
| 14 | | 36.9% | Zero-shot | Apr 2024 | Official leaderboard |
| 15 | ChatGLM3-6B | 35.5% | Zero-shot | Oct 2023 | Official leaderboard |
| 16 | Baichuan2-13B-Chat | 35.5% | Zero-shot | Sep 2023 | Official leaderboard |
| 17 | | 34.2% | Zero-shot | Aug 2024 | Official leaderboard |
| 18 | | 34.2% | Zero-shot | Feb 2024 | Official leaderboard |
| 19 | chatglm2-6b | 32.8% | Zero-shot | Jun 2023 | Official leaderboard |
| 20 | | 30.4% | Few-shot | Sep 2023 | Official leaderboard |
| 21 | | 26.6% | Chain of thought | Sep 2023 | Official leaderboard |
| 22 | | 25.4% | Chain of thought | Sep 2023 | Official leaderboard |
Rankings by part
We read 22 parts from the benchmark’s paper in 3 groups: cognitive task, domain and generation. Each ranks models on one part of the test. Qwen Plus leads 7 of them, and 5 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.
M-P
22 models we track have a score here.
- 1
Qwen Plus Apr 2025 63.8%
- 2
Qwen2-72B 62.6%
- 3
Qwen2.5-72B 60.1%
- 4
Qwen-Turbo 54.9%
- 5
Qwen1.5-72B ↑1 52.4%
- 6
Gemini 2.0 Flash Feb 2025 ↓1 52.3%
- 7
Qwen2-7B-Instruct 47.8%
- 8
Qwen1.5-14B-Chat 44.3%
- 9 GLM-4-9B-Chat 42.6%
- 10
Qwen1.5-7B-Chat ↑3 40.4%
M-R
22 models we track have a score here.
- 1
Qwen Plus Apr 2025 60.7%
- 2
Qwen2-72B 57.9%
- 3
Gemini 2.0 Flash Feb 2025 ↑2 57.4%
- 4
Qwen2.5-72B ↓1 56.8%
- 5
Qwen-Turbo ↓1 51.8%
- 6
Qwen1.5-14B-Chat ↑2 50.8%
- 7
Qwen1.5-72B ↓1 49.7%
- 8
Qwen2-7B-Instruct ↓1 47%
- 9 GLM-4-9B-Chat 45.4%
- 10
Yi-1.5-9B-Chat ↑1 44.7%
M-E
22 models we track have a score here.
- 1
Qwen-Turbo ↑3 70.5%
- 2
Qwen2-72B 62.1%
- 3
Qwen2.5-72B 60.6%
- 4
Qwen Plus Apr 2025 ↓3 58.3%
- 5
Gemini 2.0 Flash Feb 2025 55.7%
- 6
Qwen1.5-72B 54.8%
- 7
Qwen2-7B-Instruct 53.8%
- 8 GLM-4-9B-Chat ↑1 53%
- 9
Qwen1.5-14B-Chat ↓1 50.8%
- 10
Qwen1.5-7B-Chat ↑3 50.8%
M-T
22 models we track have a score here.
- 1
Qwen2.5-72B ↑2 70.5%
- 2
Qwen Plus Apr 2025 ↓1 68.2%
- 3
Qwen2-72B ↓1 66.7%
- 4
Qwen-Turbo 59.1%
- 5
Qwen1.5-14B-Chat ↑3 55.3%
- 6
Gemini 2.0 Flash Feb 2025 ↓1 54.5%
- 7
Qwen2-7B-Instruct 53.8%
- 8
phi-3-medium 14B ↑6 52.3%
- 9
Qwen1.5-72B ↓3 51.5%
- 10
Qwen1.5-7B-Chat ↑3 51.5%
U-I
22 models we track have a score here.
- 1
Qwen Plus Apr 2025 63.8%
- 2
Qwen2-72B 63.2%
- 3
Qwen2.5-72B 60.5%
- 4
Qwen-Turbo 55.6%
- 5
Gemini 2.0 Flash Feb 2025 53%
- 6
Qwen1.5-72B 51.8%
- 7
Qwen2-7B-Instruct 48%
- 8 GLM-4-9B-Chat ↑1 44.6%
- 9
Qwen-14B ↑3 41.8%
- 10
Qwen1.5-14B-Chat ↓2 41.8%
U-S
22 models we track have a score here.
- 1
Qwen2-72B ↑1 74.5%
- 2
Qwen2.5-72B ↑1 66.7%
- 3
Qwen Plus Apr 2025 ↓2 63.8%
- 4
Qwen-Turbo 63.7%
- 5
Qwen2-7B-Instruct ↑2 57.9%
- 6
Qwen1.5-72B 56.7%
- 7
Yi-1.5-9B-Chat ↑4 56.1%
- 8
InternLM2.5-20B-Chat ↑2 54.2%
- 9
phi-3-medium 14B ↑5 52.9%
- 10
Gemini 2.0 Flash Feb 2025 ↓5 52.7%
U-V
22 models we track have a score here.
- 1
Qwen2-72B ↑1 68.7%
- 2
Qwen Plus Apr 2025 ↓1 67.2%
- 3
Qwen-Turbo ↑1 64.5%
- 4
Qwen2.5-72B ↓1 62.7%
- 5
Gemini 2.0 Flash Feb 2025 62.2%
- 6
Yi-1.5-9B-Chat ↑5 58.7%
- 7
Qwen1.5-72B ↓1 56.4%
- 8
Qwen2-7B-Instruct ↓1 56.3%
- 9
InternLM2.5-20B-Chat ↑1 54.9%
- 10 GLM-4-9B-Chat ↓1 52.9%
I-D
22 models we track have a score here.
- 1
Qwen2-72B ↑1 56.2%
- 2
Gemini 2.0 Flash Feb 2025 ↑3 55.6%
- 3
Qwen-Turbo ↑1 55.3%
- 4
Qwen2.5-72B ↓1 55.1%
- 5
Qwen1.5-72B ↑1 54.8%
- 6
Qwen Plus Apr 2025 ↓5 53.9%
- 7
Qwen2-7B-Instruct 50%
- 8 GLM-4-9B-Chat ↑1 48%
- 9
Qwen-14B ↑3 47.6%
- 10
phi-3-medium 14B ↑4 45.8%
I-N
22 models we track have a score here.
- 1
Gemini 2.0 Flash Feb 2025 ↑4 66.5%
- 2
Qwen2-72B 64.2%
- 3
Qwen2.5-72B 62.6%
- 4 GLM-4-9B-Chat ↑5 56.3%
- 5
Qwen1.5-72B ↑1 56.1%
- 6
Qwen Plus Apr 2025 ↓5 54.1%
- 7
phi-3-medium 14B ↑7 51.6%
- 8
InternLM2.5-20B-Chat ↑2 51.5%
- 9
Yi-1.5-9B-Chat ↑2 50.5%
- 10
Qwen1.5-14B-Chat ↓2 49.7%
I-P
22 models we track have a score here.
- 1
Qwen2-72B ↑1 63.5%
- 2
Gemini 2.0 Flash Feb 2025 ↑3 60.5%
- 3
Qwen2.5-72B 59.7%
- 4 GLM-4-9B-Chat ↑5 59.2%
- 5
Qwen Plus Apr 2025 ↓4 59.2%
- 6
Qwen1.5-72B 59.2%
- 7
Qwen-Turbo ↓3 57.6%
- 8
Qwen2-7B-Instruct ↓1 53.4%
- 9
Qwen1.5-14B-Chat ↓1 51.3%
- 10
phi-3-medium 14B ↑4 48.5%
I-S
22 models we track have a score here.
- 1
Qwen2-72B ↑1 73.8%
- 2
Qwen2.5-72B ↑1 73.4%
- 3
Qwen-Turbo ↑1 73.3%
- 4
Qwen Plus Apr 2025 ↓3 73.3%
- 5
Gemini 2.0 Flash Feb 2025 71.1%
- 6
Qwen1.5-72B 68.9%
- 7
Qwen2-7B-Instruct 68%
- 8
InternLM2.5-20B-Chat ↑2 64.2%
- 9
Qwen1.5-14B-Chat ↓1 61.2%
- 10 GLM-4-9B-Chat ↓1 60.7%
I-G
22 models we track have a score here.
- 1
Qwen Plus Apr 2025 56.1%
- 2
Qwen2-72B 54.5%
- 3
Qwen2.5-72B 53%
- 4
Gemini 2.0 Flash Feb 2025 ↑1 52.3%
- 5
Qwen-Turbo ↓1 47.7%
- 6
Qwen1.5-72B 33%
- 7
Qwen1.5-14B-Chat ↑1 32.7%
- 8
Qwen2-7B-Instruct ↓1 31.5%
- 9
InternLM2.5-20B-Chat ↑1 29.6%
- 10
Qwen-14B ↑2 28.4%
PP
22 models we track have a score here.
- 1
Qwen Plus Apr 2025 64%
- 2
Qwen2-72B 63.5%
- 3
Qwen2.5-72B 61%
- 4
Qwen-Turbo 57.5%
- 5
Gemini 2.0 Flash Feb 2025 55.4%
- 6
Qwen1.5-72B 53.6%
- 7
Qwen2-7B-Instruct 49.5%
- 8
Qwen1.5-14B-Chat 45.5%
- 9 GLM-4-9B-Chat 44.3%
- 10
InternLM2.5-20B-Chat 42.5%
Aqua
22 models we track have a score here.
- 1
Qwen Plus Apr 2025 57.5%
- 2
Qwen2.5-72B ↑1 57%
- 3
Qwen2-72B ↓1 55.8%
- 4
Gemini 2.0 Flash Feb 2025 ↑1 54.4%
- 5
Qwen-Turbo ↓1 52.5%
- 6
Qwen2-7B-Instruct ↑1 47%
- 7
Qwen1.5-72B ↓1 46.3%
- 8
Qwen1.5-14B-Chat 44.9%
- 9 GLM-4-9B-Chat 43.2%
- 10
Yi-1.5-9B-Chat ↑1 42%
GS
22 models we track have a score here.
- 1
Qwen2.5-72B ↑2 65.1%
- 2
Gemini 2.0 Flash Feb 2025 ↑3 63.1%
- 3
Qwen Plus Apr 2025 ↓2 59.2%
- 4
Qwen-Turbo 56.9%
- 5
Qwen1.5-14B-Chat ↑3 55.6%
- 6
Qwen2-72B ↓4 55.4%
- 7
InternLM2.5-20B-Chat ↑3 48%
- 8
Qwen1.5-72B ↓2 44.5%
- 9 GLM-4-9B-Chat 44.3%
- 10
Qwen2-7B-Instruct ↓3 43.4%
TCH
22 models we track have a score here.
- 1
Qwen2-72B ↑1 61.9%
- 2
Qwen Plus Apr 2025 ↓1 61.5%
- 3
Qwen2.5-72B 57.7%
- 4
Qwen1.5-14B-Chat ↑4 57.4%
- 5
Qwen-14B ↑7 57.2%
- 6 GLM-4-9B-Chat ↑3 57.2%
- 7
Qwen1.5-72B ↓1 56.1%
- 8
Qwen-Turbo ↓4 52.7%
- 9
Yi-1.5-9B-Chat ↑2 51.3%
- 10
InternLM2.5-20B-Chat 50.8%
Fore
22 models we track have a score here.
- 1
Qwen2-72B ↑1 56.3%
- 2
Qwen Plus Apr 2025 ↓1 55.2%
- 3
Gemini 2.0 Flash Feb 2025 ↑2 55%
- 4
Qwen2.5-72B ↓1 54.4%
- 5
Qwen-Turbo ↓1 52.6%
- 6
Qwen1.5-72B 50.3%
- 7
Qwen2-7B-Instruct 46.5%
- 8 GLM-4-9B-Chat ↑1 44.6%
- 9
Yi-1.5-9B-Chat ↑2 41.3%
- 10
Qwen1.5-14B-Chat ↓2 40.1%
AST
22 models we track have a score here.
- 1
Qwen Plus Apr 2025 62.9%
- 2
Qwen2-72B 62.7%
- 3
Qwen2.5-72B 58.9%
- 4
Gemini 2.0 Flash Feb 2025 ↑1 56.7%
- 5
Qwen-Turbo ↓1 54.7%
- 6
Qwen1.5-72B 52.2%
- 7
Qwen2-7B-Instruct 46.8%
- 8
Qwen1.5-14B-Chat 43.5%
- 9
InternLM2.5-20B-Chat ↑1 43.1%
- 10 GLM-4-9B-Chat ↓1 40.9%
G-QA
ROUGE-L on a 0 to 1 scale. 22 models we track have a score here.
- 1
Qwen-Turbo ↑3 0.40
- 2
Qwen2.5-72B ↑1 0.39
- 3
Qwen Plus Apr 2025 ↓2 0.38
- 4
Qwen1.5-72B ↑2 0.37
- 5
Qwen1.5-7B-Chat ↑8 0.37
- 6
Qwen2-72B ↓4 0.37
- 7 Baichuan2-13B-Chat ↑9 0.37
- 8
Qwen1.5-14B-Chat 0.35
- 9
Yi-1.5-9B-Chat ↑2 0.34
- 10 GLM-4-9B-Chat ↓1 0.34
G-PS
ROUGE-L on a 0 to 1 scale. 22 models we track have a score here.
- 1
Qwen1.5-72B ↑5 0.35
- 2
Qwen1.5-14B-Chat ↑6 0.35
- 3
Qwen2.5-72B 0.35
- 4 Baichuan2-13B-Chat ↑12 0.35
- 5
Phi-3.5-mini ↑12 0.35
- 6
Qwen2-72B ↓4 0.35
- 7
Qwen1.5-7B-Chat ↑6 0.34
- 8
Yi-1.5-9B-Chat ↑3 0.33
- 9
Qwen-Turbo ↓5 0.33
- 10 GLM-4-9B-Chat ↓1 0.32
G-CA
ROUGE-L on a 0 to 1 scale. 22 models we track have a score here.
- 1
Qwen-Turbo ↑3 0.17
- 2
Qwen2.5-72B ↑1 0.17
- 3
Qwen Plus Apr 2025 ↓2 0.17
- 4
Qwen2-72B ↓2 0.17
- 5
Qwen1.5-72B ↑1 0.16
- 6 Baichuan2-13B-Chat ↑10 0.16
- 7
Qwen1.5-7B-Chat ↑6 0.16
- 8
Qwen1.5-14B-Chat 0.15
- 9 GLM-4-9B-Chat 0.15
- 10
Phi-3.5-mini ↑7 0.15
Generation overall
ROUGE-L on a 0 to 1 scale. 22 models we track have a score here.
- 1
Qwen-Turbo ↑3 0.21
- 2
Qwen2.5-72B ↑1 0.21
- 3
Qwen2-72B ↓1 0.21
- 4
Qwen Plus Apr 2025 ↓3 0.21
- 5
Qwen1.5-72B ↑1 0.20
- 6 Baichuan2-13B-Chat ↑10 0.20
- 7
Qwen1.5-7B-Chat ↑6 0.20
- 8
Qwen1.5-14B-Chat 0.20
- 9
Phi-3.5-mini ↑8 0.19
- 10 GLM-4-9B-Chat ↓1 0.19
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 AgriEval (Harbin Institute of Technology) paper
| Model | Effort | Harness | Score | Cost per task | Output tokens |
|---|---|---|---|---|---|
| Zero-shot | 63.2% | ||||
| Zero-shot | 62.7% | ||||
| Few-shot | 61.3% | ||||
| Zero-shot | 60.3% | ||||
| Few-shot | 59% | ||||
| Few-shot | 56.3% | ||||
| Zero-shot | 55.8% | ||||
| Zero-shot | 55.3% | ||||
| Zero-shot | 53% | ||||
| Chain of thought | 52.7% | ||||
| Few-shot | 49.4% | ||||
| Chain of thought | 49% | ||||
| Zero-shot | 48.8% | ||||
| Chain of thought | 47.8% | ||||
| Few-shot | 47.1% | ||||
| Zero-shot | 45.1% | ||||
| GLM-4-9B-Chat | Chain of thought | 43.8% | |||
| GLM-4-9B-Chat | Zero-shot | 43.7% | |||
| Chain of thought | 42.6% | ||||
| Zero-shot | 42.3% | ||||
| Chain of thought | 41.3% | ||||
| Zero-shot | 41.2% | ||||
| Chain of thought | 41.2% | ||||
| Few-shot | 41.2% | ||||
| Chain of thought | 40.5% | ||||
| Few-shot | 40.1% | ||||
| GLM-4-9B-Chat | Few-shot | 40.1% | |||
| Few-shot | 39.8% | ||||
| Chain of thought | 39.3% | ||||
| Chain of thought | 38.4% | ||||
| Zero-shot | 38.3% | ||||
| Few-shot | 37.3% | ||||
| Zero-shot | 36.9% | ||||
| Zero-shot | 36.5% | ||||
| ChatGLM3-6B | Zero-shot | 35.5% | |||
| Baichuan2-13B-Chat | Zero-shot | 35.5% | |||
| Chain of thought | 35.4% | ||||
| Baichuan2-13B-Chat | Chain of thought | 35.2% | |||
| Few-shot | 35.2% | ||||
| Few-shot | 34.4% | ||||
| Zero-shot | 34.2% | ||||
| Zero-shot | 34.2% | ||||
| Chain of thought | 33.4% | ||||
| chatglm2-6b | Zero-shot | 32.8% | |||
| chatglm2-6b | Chain of thought | 31.6% | |||
| Chain of thought | 31.4% | ||||
| chatglm2-6b | Few-shot | 30.8% | |||
| ChatGLM3-6B | Few-shot | 30.5% | |||
| Few-shot | 30.4% | ||||
| Few-shot | 29.8% | ||||
| ChatGLM3-6B | Chain of thought | 29.4% | |||
| Zero-shot | 29.1% | ||||
| Few-shot | 28.4% | ||||
| Chain of thought | 26.6% | ||||
| Zero-shot | 26.1% | ||||
| Chain of thought | 25.6% | ||||
| Chain of thought | 25.4% | ||||
| Zero-shot | 23.8% | ||||
| Few-shot | 21.5% | ||||
| Baichuan2-13B-Chat | Few-shot | 21% | |||
| Few-shot | 20.5% |
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 AgriEval (Harbin Institute of Technology) paper
| Model | Effort | Harness | Score | Cost per task | Output tokens |
|---|---|---|---|---|---|
| Zero-shot | 63.2% | ||||
| Zero-shot | 62.7% | ||||
| Few-shot | 61.3% | ||||
| Zero-shot | 60.3% | ||||
| Few-shot | 59% | ||||
| Few-shot | 56.3% | ||||
| Zero-shot | 55.8% | ||||
| Zero-shot | 55.3% | ||||
| Zero-shot | 53% | ||||
| Chain of thought | 52.7% | ||||
| Few-shot | 49.4% | ||||
| Chain of thought | 49% | ||||
| Zero-shot | 48.8% | ||||
| Chain of thought | 47.8% | ||||
| Few-shot | 47.1% | ||||
| Zero-shot | 45.1% | ||||
| GLM-4-9B-Chat | Chain of thought | 43.8% | |||
| GLM-4-9B-Chat | Zero-shot | 43.7% | |||
| Chain of thought | 42.6% | ||||
| Zero-shot | 42.3% | ||||
| Chain of thought | 41.3% | ||||
| Zero-shot | 41.2% | ||||
| Chain of thought | 41.2% | ||||
| Few-shot | 41.2% | ||||
| Chain of thought | 40.5% | ||||
| Few-shot | 40.1% | ||||
| GLM-4-9B-Chat | Few-shot | 40.1% | |||
| Few-shot | 39.8% | ||||
| Chain of thought | 39.3% | ||||
| Chain of thought | 38.4% | ||||
| Zero-shot | 38.3% | ||||
| Few-shot | 37.3% | ||||
| Zero-shot | 36.9% | ||||
| Zero-shot | 36.5% | ||||
| ChatGLM3-6B | Zero-shot | 35.5% | |||
| Baichuan2-13B-Chat | Zero-shot | 35.5% | |||
| Chain of thought | 35.4% | ||||
| Baichuan2-13B-Chat | Chain of thought | 35.2% | |||
| Few-shot | 35.2% | ||||
| Few-shot | 34.4% | ||||
| Zero-shot | 34.2% | ||||
| Zero-shot | 34.2% | ||||
| Chain of thought | 33.4% | ||||
| chatglm2-6b | Zero-shot | 32.8% | |||
| chatglm2-6b | Chain of thought | 31.6% | |||
| Chain of thought | 31.4% | ||||
| chatglm2-6b | Few-shot | 30.8% | |||
| ChatGLM3-6B | Few-shot | 30.5% | |||
| Few-shot | 30.4% | ||||
| Few-shot | 29.8% | ||||
| ChatGLM3-6B | Chain of thought | 29.4% | |||
| Zero-shot | 29.1% | ||||
| Few-shot | 28.4% | ||||
| Chain of thought | 26.6% | ||||
| Zero-shot | 26.1% | ||||
| Chain of thought | 25.6% | ||||
| Chain of thought | 25.4% | ||||
| Zero-shot | 23.8% | ||||
| Few-shot | 21.5% | ||||
| Baichuan2-13B-Chat | Few-shot | 21% | |||
| Few-shot | 20.5% |
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: The AgriEval (Harbin Institute of Technology) 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 AgriEval (Harbin Institute of Technology) 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.
More in Knowledge
- MMLU-Pro In use Harder 10-option version of MMLU with reasoning-focused questions across 14 subjects.
Intern-S2-397B 89.8%
- MMLU In use 57-subject multiple-choice exam questions from elementary to professional level. Scored by accuracy.
Seed1.8 92.3%
- SuperGPQA In use 26,529 graduate-level multiple-choice questions across 285 disciplines. Accuracy.
Qwen 3.6 Max (Preview) 73.9%
- WideSearch In use 200 bilingual broad info-seeking tasks: agent collects many atomic facts into a table. Scored by success rate and row/item F1.
Tencent Hy4 preview 83.9%
- BBEH micro average In use Models solve harder replacements for each of the 23 BIG-Bench Hard tasks, 4,520 examples in total. The score is the share of all examples answered correctly.
Gemma 4 31B IT 74.4%
- HLE-Diamond In use Models answer 1,000 cleaned expert questions from Humanity's Last Exam, 500 on reasoning and 500 on knowledge, closed book. Graded on exact answers. Claude Fable 5.2 77.9%
Data audit
9 of 10 checks pass, 1 warning. Look first at the Newest score check. The newest came 19 months ago, in April 2025, for Qwen Plus. Every benchmark
Catalog record 2 pass
- Pass Task access All tasks are public.
- Pass Confidence Checked against a primary source.
Scores 1 warning 3 pass 1 note
- Pass Independent runs Scores from the benchmark’s paper (51 entries).
- Pass Models scored 22 models we track, plus 29 other entries on its leaderboard.
- Warning Newest score The newest came 19 months ago, in April 2025, for Qwen Plus.
- Note Frontier models None of the top 10 has a score. The strongest tested is Gemini 2.0 Flash (#147).
- Pass Headroom The best is 63.2%, under 85%, so it still separates models.
Leaderboard import 3 pass
- Pass Importer We read the board on arxiv.org: 61 rows for models we track and 29 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.