K-MetBench
Models answer 1,774 questions from the Korean National Meteorological Engineer Examination under a zero-shot protocol. The benchmark reports answer accuracy across five subjects and separate visual and Korean-specific subsets. Expert rationales support an additional rubric score for reasoning.
The paper evaluates 55 models. The opened leaderboard data contains 71 rows, including separate thinking settings. The 82 visual questions, 141 reasoning questions, and 73 Korean-specific questions are overlapping diagnostic subsets.
- Made by
- KAIST, INEEJI and Kongju National University
- Released
- April 2026
- Kind
- Science questions
- Level
- Expert
- Scored
- Checked automatically
- Answers
- Multiple choice
- Tasks
- All public
- Measures
- Capability
- Subject
- Earth and environmental sciences
- Language
- Korean
- Interaction
- Single turn
- Items from
- Exams and contests
- Refresh
- One fixed set
- Run by
- Anyone
- Occupations
- Architecture and engineering
- In views
- Vision
- Strongest model tested
- Kimi K3 #25 on the capability index
- Last model added
- July 2026
- Scores refreshed
- November 20th, 2026
In the authors’ words
The development of practical (multimodal) large language model assistants for Korean weather forecasters is hindered by the absence of a multidimensional, expert-level evaluation framework grounded in authoritative sources. To address this, we introduce K-MetBench, a diagnostic benchmark grounded in national qualification exams. It exposes critical gaps across four dimensions: expert visual reasoning of charts, logical validity via expert-verified rationales, Korean-specific geo-cultural comprehension, and fine-grained domain analysis. Our evaluation of 55 models reveals a profound modality gap in interpreting specialized diagrams and a reasoning gap where models hallucinate logic despite correct predictions. Crucially, Korean models outperform significantly larger global models in local contexts, demonstrating that parameter scaling alone cannot resolve cultural dependencies. K-MetBench serves as a roadmap for developing reliable, culturally aware expert AI agents. The dataset is available at https://huggingface.co/datasets/soyeonbot/K-MetBench .
Gemini 3 Pro leads with 93.7% among models released by July 2026 Gemini 3 Pro leads with 93.7% among models released by July 2026
Best score per entry from the benchmark's own leaderboard, read on November 20th, 2026. 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 July 2026, 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
- 93.7
Gemini 3 Pro
- 87.8 GPT-5.2
- 86.5
Gemma 4 31B IT
- 84.4
Qwen3-VL-235B-A22B
- 83.0
Qwen3.5-27B
- 82.9
Qwen 3.6 35B-A3B
- 82.6
DeepSeek-V4-Pro
- 82.5 Kimi K2.6
- 81.8
Qwen3.5-35B-A3B
- 81.5 Kimi K3
- 80.1
DeepSeek V4 Flash 0731
- 79.6
DeepSeek-V4-Flash
- 78.6
Qwen3-VL-32B-Thinking
- 77.8
Command A Reasoning
- 77.3 gpt-oss-120b
- 76.7
Qwen3-30B-A3B-Thinking
- 76.1 A.X-4.0
- 75.9
EXAONE 4.5 33B
- 74.9
Qwen3-VL-30B-A3B-Thinking
- 74.9
Qwen3.5-9B
- 73.7
Qwen3-14B
- 72.4
Qwen3-VL-235B-A22B-Instruct
- 71.7
Qwen3-VL-8B-Thinking
- 71.5 gpt-oss-20b
- 70.1
Qwen3-8B
- 67.8
Qwen3-4B-Thinking-2507
- 67.5
Qwen3-VL-32B-Instruct
- 67.1
Qwen2.5-VL-72B
- 66.1
Qwen3-VL-4B-Thinking
- 65.5 c4ai-command-a-03-2025
- 64.7
Qwen3-30B-A3B-Instruct
- 62.2
Qwen3-VL-30B-A3B-Instruct
- 60.1
Qwen2.5-VL-32B
- 59.9 Llama-3.1-70B-Instruct
- 59.9
EXAONE 4.0 32B
- 58.7 VARCO-Vision-2.0-14B
- 57.3 InternVL3.5-38B-Instruct
- 56.9
Llama 3.2 90B
- 55.7 A.X-4.0-Light
- 53.8
Qwen3-VL-8B-Instruct
- 52.5 A.X-4.0-VL-Light
- 51.5 qwen3-4b-instruct-2507
- 51.5
Phi-4
- 51.0
Qwen3-VL-4B-Instruct
- 50.8 HyperCLOVAX-SEED-Think-14B
- 47.9 InternVL3.5-14B-Instruct
- 47.5
Qwen3-32B
- 46.8
Qwen3-1.7B
- 46.1 InternVL3.5-8B-Instruct
- 46.0
Qwen2.5-VL-7B
- 41.8 Llama-3.1-8B-Instruct
- 41.5 InternVL3.5-4B-Instruct
- 40.9
Qwen2.5-VL-3B
- 37.4 EXAONE-4.0-1.2B
- 35.2 VARCO-Vision-2.0-1.7B
- 33.8 Llama-3.2-3B-Instruct
- 32.1
Qwen3-0.6B
- 32.0 HyperCLOVAX-SEED-Vision-Instruct-3B
- 31.0 InternVL3.5-2B-Instruct
- 30.6 HyperCLOVAX-SEED-Text-Instruct-1.5B
- 30.4 Phi-4-mini-instruct
- 23.8 InternVL3.5-1B-Instruct
- 13.2 HyperCLOVAX-SEED-Text-Instruct-0.5B
- 12.6
Phi-4-mini-reasoning
- 3.55 Llama-3.2-1B-Instruct
- 1
Gemini 3 Pro thinking 93.7%
- 2 GPT-5.2 thinking 87.8%
- 3
Gemma 4 31B IT thinking 86.5%
- 4
Qwen3-VL-235B-A22B thinking 84.4%
- 5
Qwen3.5-27B thinking 83%
- 6
Qwen 3.6 35B-A3B thinking 82.9%
- 7
DeepSeek-V4-Pro 82.6%
- 8 Kimi K2.6 82.5%
- 9
Qwen3.5-35B-A3B thinking 81.8%
- 10 Kimi K3 81.5%
- 11
DeepSeek V4 Flash 0731 80.1%
- 12
DeepSeek-V4-Flash 79.6%
- 13
Qwen3-VL-32B-Thinking 78.6%
- 14
Command A Reasoning 77.8%
- 15 gpt-oss-120b 77.3%
- 16
Qwen3-30B-A3B-Thinking Jul 2025 76.7%
- 17 A.X-4.0 76.1%
- 18
EXAONE 4.5 33B 75.9%
- 19
Qwen3-VL-30B-A3B-Thinking 74.9%
- 20
Qwen3.5-9B thinking 74.9%
- 21
Qwen3-14B 73.7%
- 22
Qwen3-VL-235B-A22B-Instruct 72.4%
- 23
Qwen3-VL-8B-Thinking 71.7%
- 24 gpt-oss-20b 71.5%
- 25
Qwen3-8B 70.1%
- 26
Qwen3-4B-Thinking-2507 67.8%
- 27
Qwen3-VL-32B-Instruct 67.5%
- 28
Qwen2.5-VL-72B 67.1%
- 29
Qwen3-VL-4B-Thinking 66.1%
- 30 c4ai-command-a-03-2025 65.5%
- 31
Qwen3-30B-A3B-Instruct Jul 2025 64.7%
- 32
Qwen3-VL-30B-A3B-Instruct 62.2%
- 33
Qwen2.5-VL-32B 60.1%
- 34 Llama-3.1-70B-Instruct 59.9%
- 35
EXAONE 4.0 32B 59.9%
- 36 VARCO-Vision-2.0-14B 58.7%
- 37 InternVL3.5-38B-Instruct 57.3%
- 38
Llama 3.2 90B 56.9%
- 39 A.X-4.0-Light 55.7%
- 40
Qwen3-VL-8B-Instruct 53.8%
- 41 A.X-4.0-VL-Light 52.5%
- 42 qwen3-4b-instruct-2507 51.5%
- 43
Phi-4 51.5%
- 44
Qwen3-VL-4B-Instruct 51%
- 45 HyperCLOVAX-SEED-Think-14B 50.8%
- 46 InternVL3.5-14B-Instruct 47.9%
- 47
Qwen3-32B 47.5%
- 48
Qwen3-1.7B 46.8%
- 49 InternVL3.5-8B-Instruct 46.1%
- 50
Qwen2.5-VL-7B 46%
- 51 Llama-3.1-8B-Instruct 41.8%
- 52 InternVL3.5-4B-Instruct 41.5%
- 53
Qwen2.5-VL-3B 40.9%
- 54 EXAONE-4.0-1.2B 37.4%
- 55 VARCO-Vision-2.0-1.7B 35.2%
- 56 Llama-3.2-3B-Instruct 33.8%
- 57
Qwen3-0.6B 32.1%
- 58 HyperCLOVAX-SEED-Vision-Instruct-3B 32%
- 59 InternVL3.5-2B-Instruct 31%
- 60 HyperCLOVAX-SEED-Text-Instruct-1.5B 30.6%
- 61 Phi-4-mini-instruct 30.4%
- 62 InternVL3.5-1B-Instruct 23.8%
- 63 HyperCLOVAX-SEED-Text-Instruct-0.5B 13.2%
- 64
Phi-4-mini-reasoning 12.6%
- 65 Llama-3.2-1B-Instruct 3.55%
Source: Official K-MetBench leaderboard
- 93.7
Gemini 3 Pro
- 87.8 GPT-5.2
- 86.5
Gemma 4 31B IT
- 84.4
Qwen3-VL-235B-A22B
- 83.0
Qwen3.5-27B
- 82.9
Qwen 3.6 35B-A3B
- 82.6
DeepSeek-V4-Pro
- 82.5 Kimi K2.6
- 81.8
Qwen3.5-35B-A3B
- 81.5 Kimi K3
- 80.1
DeepSeek V4 Flash 0731
- 79.6
DeepSeek-V4-Flash
- 78.6
Qwen3-VL-32B-Thinking
- 77.8
Command A Reasoning
- 77.3 gpt-oss-120b
- 76.7
Qwen3-30B-A3B-Thinking
- 75.9
EXAONE 4.5 33B
- 74.9
Qwen3-VL-30B-A3B-Thinking
- 74.9
Qwen3.5-9B
- 73.7
Qwen3-14B
- 72.4
Qwen3-VL-235B-A22B-Instruct
- 71.7
Qwen3-VL-8B-Thinking
- 71.5 gpt-oss-20b
- 70.1
Qwen3-8B
- 67.8
Qwen3-4B-Thinking-2507
- 67.5
Qwen3-VL-32B-Instruct
- 67.1
Qwen2.5-VL-72B
- 66.1
Qwen3-VL-4B-Thinking
- 64.7
Qwen3-30B-A3B-Instruct
- 62.2
Qwen3-VL-30B-A3B-Instruct
- 60.1
Qwen2.5-VL-32B
- 59.9
EXAONE 4.0 32B
- 56.9
Llama 3.2 90B
- 53.8
Qwen3-VL-8B-Instruct
- 51.5 qwen3-4b-instruct-2507
- 51.5
Phi-4
- 51.0
Qwen3-VL-4B-Instruct
- 47.5
Qwen3-32B
- 46.8
Qwen3-1.7B
- 46.0
Qwen2.5-VL-7B
- 40.9
Qwen2.5-VL-3B
- 32.1
Qwen3-0.6B
- 12.6
Phi-4-mini-reasoning
- 1
Gemini 3 Pro thinking 93.7%
- 2 GPT-5.2 thinking 87.8%
- 3
Gemma 4 31B IT thinking 86.5%
- 4
Qwen3-VL-235B-A22B thinking 84.4%
- 5
Qwen3.5-27B thinking 83%
- 6
Qwen 3.6 35B-A3B thinking 82.9%
- 7
DeepSeek-V4-Pro 82.6%
- 8 Kimi K2.6 82.5%
- 9
Qwen3.5-35B-A3B thinking 81.8%
- 10 Kimi K3 81.5%
- 11
DeepSeek V4 Flash 0731 80.1%
- 12
DeepSeek-V4-Flash 79.6%
- 13
Qwen3-VL-32B-Thinking 78.6%
- 14
Command A Reasoning 77.8%
- 15 gpt-oss-120b 77.3%
- 16
Qwen3-30B-A3B-Thinking Jul 2025 76.7%
- 17
EXAONE 4.5 33B 75.9%
- 18
Qwen3-VL-30B-A3B-Thinking 74.9%
- 19
Qwen3.5-9B thinking 74.9%
- 20
Qwen3-14B 73.7%
- 21
Qwen3-VL-235B-A22B-Instruct 72.4%
- 22
Qwen3-VL-8B-Thinking 71.7%
- 23 gpt-oss-20b 71.5%
- 24
Qwen3-8B 70.1%
- 25
Qwen3-4B-Thinking-2507 67.8%
- 26
Qwen3-VL-32B-Instruct 67.5%
- 27
Qwen2.5-VL-72B 67.1%
- 28
Qwen3-VL-4B-Thinking 66.1%
- 29
Qwen3-30B-A3B-Instruct Jul 2025 64.7%
- 30
Qwen3-VL-30B-A3B-Instruct 62.2%
- 31
Qwen2.5-VL-32B 60.1%
- 32
EXAONE 4.0 32B 59.9%
- 33
Llama 3.2 90B 56.9%
- 34
Qwen3-VL-8B-Instruct 53.8%
- 35 qwen3-4b-instruct-2507 51.5%
- 36
Phi-4 51.5%
- 37
Qwen3-VL-4B-Instruct 51%
- 38
Qwen3-32B 47.5%
- 39
Qwen3-1.7B 46.8%
- 40
Qwen2.5-VL-7B 46%
- 41
Qwen2.5-VL-3B 40.9%
- 42
Qwen3-0.6B 32.1%
- 43
Phi-4-mini-reasoning 12.6%
Source: Official K-MetBench leaderboard
| # | Entry | Score | Setting | Released | Run on | Source |
|---|---|---|---|---|---|---|
| 1 | | 93.7% | thinking | Nov 2025 | Dec 2025 | Official leaderboard |
| 2 | GPT-5.2 | 87.8% | thinking | Dec 2025 | Dec 2025 | Official leaderboard |
| 3 | | 86.5% | thinking | Apr 2026 | Jun 2026 | Official leaderboard |
| 4 | | 84.4% | thinking | Sep 2025 | Oct 2025 | Official leaderboard |
| 5 | | 83% | thinking | Feb 2026 | May 2026 | Official leaderboard |
| 6 | | 82.9% | thinking | Apr 2026 | May 2026 | Official leaderboard |
| 7 | | 82.6% | Apr 2026 | Aug 2026 | Official leaderboard | |
| 8 | Kimi K2.6 | 82.5% | Apr 2026 | Aug 2026 | Official leaderboard | |
| 9 | | 81.8% | thinking | Feb 2026 | May 2026 | Official leaderboard |
| 10 | Kimi K3 | 81.5% | Jul 2026 | Aug 2026 | Official leaderboard | |
| 11 | | 80.1% | Jul 2026 | Aug 2026 | Official leaderboard | |
| 12 | | 79.6% | Apr 2026 | Aug 2026 | Official leaderboard | |
| 13 | | 78.6% | Oct 2025 | Oct 2025 | Official leaderboard | |
| 14 | | 77.8% | Aug 2025 | Oct 2025 | Official leaderboard | |
| 15 | gpt-oss-120b | 77.3% | Aug 2025 | Oct 2025 | Official leaderboard | |
| 16 | | 76.7% | Jul 2025 | Dec 2025 | Official leaderboard | |
| 17 | A.X-4.0 72B | 76.1% | Oct 2025 | Official leaderboard | ||
| 18 | | 75.9% | Apr 2026 | Apr 2026 | Official leaderboard | |
| 19 | | 74.9% | Oct 2025 | Dec 2025 | Official leaderboard | |
| 20 | | 74.9% | thinking | Feb 2026 | May 2026 | Official leaderboard |
| 21 | | 73.7% | Apr 2025 | Oct 2025 | Official leaderboard | |
| 22 | | 72.4% | Sep 2025 | Oct 2025 | Official leaderboard | |
| 23 | | 71.7% | Oct 2025 | Oct 2025 | Official leaderboard | |
| 24 | gpt-oss-20b | 71.5% | Aug 2025 | Oct 2025 | Official leaderboard | |
| 25 | | 70.1% | Apr 2025 | Oct 2025 | Official leaderboard | |
| 26 | | 67.8% | Aug 2025 | Oct 2025 | Official leaderboard | |
| 27 | | 67.5% | Oct 2025 | Oct 2025 | Official leaderboard | |
| 28 | | 67.1% | Jan 2025 | Dec 2025 | Official leaderboard | |
| 29 | | 66.1% | Oct 2025 | Oct 2025 | Official leaderboard | |
| 30 | c4ai-command-a-03-2025 111B | 65.5% | Oct 2025 | Official leaderboard | ||
| 31 | | 64.7% | Jul 2025 | Oct 2025 | Official leaderboard | |
| 32 | | 62.2% | Oct 2025 | Oct 2025 | Official leaderboard | |
| 33 | | 60.1% | Mar 2025 | Dec 2025 | Official leaderboard | |
| 34 | Llama-3.1-70B-Instruct 71B | 59.9% | Dec 2025 | Official leaderboard | ||
| 35 | | 59.9% | Jul 2025 | Oct 2025 | Official leaderboard | |
| 36 | VARCO-Vision-2.0-14B 14B | 58.7% | Oct 2025 | Official leaderboard | ||
| 37 | InternVL3.5-38B-Instruct 38B | 57.3% | Oct 2025 | Official leaderboard | ||
| 38 | | 56.9% | Sep 2024 | Oct 2025 | Official leaderboard | |
| 39 | A.X-4.0-Light 7B | 55.7% | Oct 2025 | Official leaderboard | ||
| 40 | | 53.8% | Oct 2025 | Oct 2025 | Official leaderboard | |
| 41 | A.X-4.0-VL-Light 8B | 52.5% | Oct 2025 | Official leaderboard | ||
| 42 | qwen3-4b-instruct-2507 | 51.5% | Aug 2025 | Oct 2025 | Official leaderboard | |
| 43 | | 51.5% | Dec 2024 | Dec 2025 | Official leaderboard | |
| 44 | | 51% | Oct 2025 | Oct 2025 | Official leaderboard | |
| 45 | HyperCLOVAX-SEED-Think-14B 15B | 50.8% | Dec 2025 | Official leaderboard | ||
| 46 | InternVL3.5-14B-Instruct 15B | 47.9% | Oct 2025 | Official leaderboard | ||
| 47 | | 47.5% | Apr 2025 | Oct 2025 | Official leaderboard | |
| 48 | | 46.8% | Apr 2025 | Oct 2025 | Official leaderboard | |
| 49 | InternVL3.5-8B-Instruct 8B | 46.1% | Oct 2025 | Official leaderboard | ||
| 50 | | 46% | Jan 2025 | Dec 2025 | Official leaderboard | |
| 51 | Llama-3.1-8B-Instruct 8B | 41.8% | Oct 2025 | Official leaderboard | ||
| 52 | InternVL3.5-4B-Instruct 5B | 41.5% | Oct 2025 | Official leaderboard | ||
| 53 | | 40.9% | Jan 2025 | Dec 2025 | Official leaderboard | |
| 54 | EXAONE-4.0-1.2B 1.2B | 37.4% | Oct 2025 | Official leaderboard | ||
| 55 | VARCO-Vision-2.0-1.7B 1.7B | 35.2% | Oct 2025 | Official leaderboard | ||
| 56 | Llama-3.2-3B-Instruct 3B | 33.8% | Oct 2025 | Official leaderboard | ||
| 57 | | 32.1% | Apr 2025 | Oct 2025 | Official leaderboard | |
| 58 | HyperCLOVAX-SEED-Vision-Instruct-3B 3B | 32% | Oct 2025 | Official leaderboard | ||
| 59 | InternVL3.5-2B-Instruct 2B | 31% | Oct 2025 | Official leaderboard | ||
| 60 | HyperCLOVAX-SEED-Text-Instruct-1.5B 1.5B | 30.6% | Dec 2025 | Official leaderboard | ||
| 61 | Phi-4-mini-instruct 4B | 30.4% | Oct 2025 | Official leaderboard | ||
| 62 | InternVL3.5-1B-Instruct 1B | 23.8% | Oct 2025 | Official leaderboard | ||
| 63 | HyperCLOVAX-SEED-Text-Instruct-0.5B 0.5B | 13.2% | Oct 2025 | Official leaderboard | ||
| 64 | | 12.6% | Apr 2025 | Dec 2025 | Official leaderboard | |
| 65 | Llama-3.2-1B-Instruct 1B | 3.55% | Dec 2025 | Official leaderboard |
| # | Model | Score | Setting | Released | Run on | Source |
|---|---|---|---|---|---|---|
| 1 | | 93.7% | thinking | Nov 2025 | Dec 2025 | Official leaderboard |
| 2 | GPT-5.2 | 87.8% | thinking | Dec 2025 | Dec 2025 | Official leaderboard |
| 3 | | 86.5% | thinking | Apr 2026 | Jun 2026 | Official leaderboard |
| 4 | | 84.4% | thinking | Sep 2025 | Oct 2025 | Official leaderboard |
| 5 | | 83% | thinking | Feb 2026 | May 2026 | Official leaderboard |
| 6 | | 82.9% | thinking | Apr 2026 | May 2026 | Official leaderboard |
| 7 | | 82.6% | Apr 2026 | Aug 2026 | Official leaderboard | |
| 8 | Kimi K2.6 | 82.5% | Apr 2026 | Aug 2026 | Official leaderboard | |
| 9 | | 81.8% | thinking | Feb 2026 | May 2026 | Official leaderboard |
| 10 | Kimi K3 | 81.5% | Jul 2026 | Aug 2026 | Official leaderboard | |
| 11 | | 80.1% | Jul 2026 | Aug 2026 | Official leaderboard | |
| 12 | | 79.6% | Apr 2026 | Aug 2026 | Official leaderboard | |
| 13 | | 78.6% | Oct 2025 | Oct 2025 | Official leaderboard | |
| 14 | | 77.8% | Aug 2025 | Oct 2025 | Official leaderboard | |
| 15 | gpt-oss-120b | 77.3% | Aug 2025 | Oct 2025 | Official leaderboard | |
| 16 | | 76.7% | Jul 2025 | Dec 2025 | Official leaderboard | |
| 17 | | 75.9% | Apr 2026 | Apr 2026 | Official leaderboard | |
| 18 | | 74.9% | Oct 2025 | Dec 2025 | Official leaderboard | |
| 19 | | 74.9% | thinking | Feb 2026 | May 2026 | Official leaderboard |
| 20 | | 73.7% | Apr 2025 | Oct 2025 | Official leaderboard | |
| 21 | | 72.4% | Sep 2025 | Oct 2025 | Official leaderboard | |
| 22 | | 71.7% | Oct 2025 | Oct 2025 | Official leaderboard | |
| 23 | gpt-oss-20b | 71.5% | Aug 2025 | Oct 2025 | Official leaderboard | |
| 24 | | 70.1% | Apr 2025 | Oct 2025 | Official leaderboard | |
| 25 | | 67.8% | Aug 2025 | Oct 2025 | Official leaderboard | |
| 26 | | 67.5% | Oct 2025 | Oct 2025 | Official leaderboard | |
| 27 | | 67.1% | Jan 2025 | Dec 2025 | Official leaderboard | |
| 28 | | 66.1% | Oct 2025 | Oct 2025 | Official leaderboard | |
| 29 | | 64.7% | Jul 2025 | Oct 2025 | Official leaderboard | |
| 30 | | 62.2% | Oct 2025 | Oct 2025 | Official leaderboard | |
| 31 | | 60.1% | Mar 2025 | Dec 2025 | Official leaderboard | |
| 32 | | 59.9% | Jul 2025 | Oct 2025 | Official leaderboard | |
| 33 | | 56.9% | Sep 2024 | Oct 2025 | Official leaderboard | |
| 34 | | 53.8% | Oct 2025 | Oct 2025 | Official leaderboard | |
| 35 | qwen3-4b-instruct-2507 | 51.5% | Aug 2025 | Oct 2025 | Official leaderboard | |
| 36 | | 51.5% | Dec 2024 | Dec 2025 | Official leaderboard | |
| 37 | | 51% | Oct 2025 | Oct 2025 | Official leaderboard | |
| 38 | | 47.5% | Apr 2025 | Oct 2025 | Official leaderboard | |
| 39 | | 46.8% | Apr 2025 | Oct 2025 | Official leaderboard | |
| 40 | | 46% | Jan 2025 | Dec 2025 | Official leaderboard | |
| 41 | | 40.9% | Jan 2025 | Dec 2025 | Official leaderboard | |
| 42 | | 32.1% | Apr 2025 | Oct 2025 | Official leaderboard | |
| 43 | | 12.6% | Apr 2025 | Dec 2025 | Official leaderboard |
Rankings by part
We read 9 parts from the benchmark’s own leaderboard in 3 groups: split, modality and domain. Each ranks models on one part of the test. Gemini 3 Pro leads every one of them. Arrows show how far a model moved from its overall rank among the models we track.
Korean specific
43 models we track have a score here.
- 1
Gemini 3 Pro 90.4%
- 2 Kimi K3 ↑8 83.6%
- 3
Gemma 4 31B IT 82.2%
- 4 GPT-5.2 ↓2 80.8%
- 5
DeepSeek V4 Flash 0731 ↑6 78.9%
- 6
DeepSeek-V4-Flash ↑6 76.1%
- 7
DeepSeek-V4-Pro 76.1%
- 8
Command A Reasoning ↑6 74.6%
- 9
Qwen3-VL-235B-A22B-Instruct ↑12 74%
- 10
Qwen3-VL-235B-A22B ↓6 72.6%
Text
43 models we track have a score here.
- 1
Gemini 3 Pro 94.6%
- 2 GPT-5.2 90.6%
- 3
Gemma 4 31B IT 87.5%
- 4
Qwen3-VL-235B-A22B 86.2%
- 5
Qwen 3.6 35B-A3B ↑1 84.8%
- 6 Kimi K2.6 ↑2 84.5%
- 7
Qwen3.5-27B ↓2 84.3%
- 8 Kimi K3 ↑2 83.2%
- 9
Qwen3.5-35B-A3B 83.1%
- 10
DeepSeek-V4-Pro ↓3 82.6%
Text only
43 models we track have a score here.
- 1
Gemini 3 Pro 94.6%
- 2 GPT-5.2 90.6%
- 3
Gemma 4 31B IT 87.5%
- 4
Qwen3-VL-235B-A22B 86.2%
- 5
Qwen 3.6 35B-A3B ↑1 84.8%
- 6 Kimi K2.6 ↑2 84.5%
- 7
Qwen3.5-27B ↓2 84.3%
- 8 Kimi K3 ↑2 83.2%
- 9
Qwen3.5-35B-A3B 83.1%
- 10
DeepSeek-V4-Pro ↓3 82.6%
Multimodal
25 models we track have a score here.
- 1
Gemini 3 Pro 75.6%
- 2
Gemma 4 31B IT ↑1 67.1%
- 3
Qwen3.5-27B ↑2 54.9%
- 4
Qwen3.5-35B-A3B ↑5 54.9%
- 5
Qwen3-VL-32B-Thinking ↑8 51.2%
- 6 GPT-5.2 ↓4 50%
- 7
Qwen3-VL-235B-A22B ↓3 48.8%
- 8
Qwen3-VL-4B-Instruct ↑29 48.8%
- 9
Qwen3.5-9B ↑10 48.8%
- 10
Qwen3-VL-4B-Thinking ↑18 47.6%
Weather analysis and forecast theory
43 models we track have a score here.
- 1
Gemini 3 Pro 92.5%
- 2 GPT-5.2 86.3%
- 3 Kimi K3 ↑7 85.8%
- 4
DeepSeek-V4-Pro ↑3 83.2%
- 5
Qwen3-VL-235B-A22B ↓1 81.5%
- 6 Kimi K2.6 ↑2 81.5%
- 7
Qwen3.5-27B ↓2 81%
- 8
Gemma 4 31B IT ↓5 81%
- 9
Qwen 3.6 35B-A3B ↓3 79.4%
- 10
Qwen3.5-35B-A3B ↓1 77.2%
Meteorological observation methods
43 models we track have a score here.
- 1
Gemini 3 Pro 97.9%
- 2 GPT-5.2 93.4%
- 3
Gemma 4 31B IT 92.8%
- 4
Qwen 3.6 35B-A3B ↑2 89.8%
- 5 Kimi K2.6 ↑3 89.2%
- 6
Qwen3-VL-235B-A22B ↓2 88.6%
- 7
DeepSeek-V4-Flash ↑5 87.4%
- 8
Qwen3.5-27B ↓3 87%
- 9
DeepSeek V4 Flash 0731 ↑2 86.8%
- 10
Qwen3.5-35B-A3B ↓1 86.4%
Atmospheric dynamics
43 models we track have a score here.
- 1
Gemini 3 Pro 94.2%
- 2
Gemma 4 31B IT ↑1 90%
- 3 GPT-5.2 ↓1 88%
- 4
Qwen3.5-27B ↑1 87.5%
- 5
Qwen3-VL-235B-A22B ↓1 87.2%
- 6
Qwen 3.6 35B-A3B 85.2%
- 7
Qwen3.5-35B-A3B ↑2 84.1%
- 8 Kimi K2.6 80.8%
- 9
Qwen3-VL-32B-Thinking ↑4 78.8%
- 10
Qwen3.5-9B ↑9 78.8%
Climatology
43 models we track have a score here.
- 1
Gemini 3 Pro 92.8%
- 2
Gemma 4 31B IT ↑1 86.4%
- 3 GPT-5.2 ↓1 86.2%
- 4 Kimi K3 ↑6 86.2%
- 5
Qwen3-VL-235B-A22B ↓1 83.2%
- 6
Qwen3.5-35B-A3B ↑3 83%
- 7
DeepSeek-V4-Pro 82.6%
- 8 Kimi K2.6 81.6%
- 9
Qwen 3.6 35B-A3B ↓3 81.4%
- 10
DeepSeek V4 Flash 0731 ↑1 81.3%
Atmospheric physics
43 models we track have a score here.
- 1
Gemini 3 Pro 91.6%
- 2 GPT-5.2 85.3%
- 3
DeepSeek-V4-Pro ↑4 84.4%
- 4
Gemma 4 31B IT ↓1 82.9%
- 5
Qwen3-VL-235B-A22B ↓1 82%
- 6 Kimi K3 ↑4 80.5%
- 7
DeepSeek V4 Flash 0731 ↑4 80.5%
- 8 Kimi K2.6 79.9%
- 9
DeepSeek-V4-Flash ↑3 79.8%
- 10
Qwen 3.6 35B-A3B ↓4 79.3%
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: Official K-MetBench leaderboard
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: Official K-MetBench leaderboard
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: Official K-MetBench leaderboard
Gemini 3 Pro on the official leaderboard passed 85% in November 2025
Models released before it came out already scored that high. All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official K-MetBench 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 Science and reasoning
- GPQA Diamond In use 198 graduate-level 'Google-proof' multiple-choice questions in biology, chemistry and physics.
Gemini 4 Carbon 97%
- Terminal-Bench-Science In use 70 scientific research-workflow tasks across five science areas, run in Harbor containers. Hidden tests give the resolution rate. GPT-6 Astra 68.1%
- BioMysteryBench In use Bioinformatics problems on anonymized real datasets that require real analysis with tools. Rubric-graded, with human-solvable and human-difficult splits. Claude Fable 5.2 83.7%
- FrontierScience In use Models solve physics, chemistry, and biology problems in an Olympiad track graded by short answer and a Research track graded by 10-point rubrics.
Intern-S2-397B 87%
- HiPhO In use Models solve theory problems from 2024 and 2025 physics olympiad exams, some with diagrams. Official marking schemes grade answers and steps, mapped to medal levels.
Gemini 3 Pro 78.3%
- LABBench2 In use 11 biology research sub-tasks (literature, figures, tables, patents, trials and more) with tool access. Macro-averaged accuracy. GPT-5.2 60.9%
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 official leaderboard (65 entries).
- Pass Models scored 43 models we track, plus 22 other entries on its leaderboard.
- Pass Newest score The newest is from July 2026, for DeepSeek V4 Flash 0731.
- Note Frontier models None of the top 10 has a score, which is fine for a retired test. The strongest tested is Kimi K3 (#25).
- Note Headroom Saturated. The best is 93.7%, past 85% since November 2025.
Leaderboard import 3 pass
- Pass Importer We read the board on kmetbench.github.io: 49 rows for models we track and 22 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.