IndicParam
Models answer multiple-choice questions from Indian competitive exams. The benchmark tests knowledge in low-resource Indic languages.
- Made by
- Ayush Maheshwari and colleagues
- Released
- November 2025
- Kind
- Knowledge in many languages
- Level
- University
- Scored
- Checked automatically
- Answers
- Multiple choice
- Tasks
- All public
- Measures
- Capability
- Subject
- Languages and literature
- Language
- Many languages
- Interaction
- Single turn
- Items from
- Exams and contests
- Refresh
- One fixed set
- Run by
- Anyone
- In views
- Many languages
- Strongest model tested
- GPT-5 #71 on the capability index
- Last model added
- October 2025
- Scores refreshed
- November 20th, 2026
Gemini-2.5-flash (snapshot unspecified) tops the leaderboard with 58% GPT-5 leads with 45% among models released by October 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. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders. The best model we track is GPT-5, at 45%. This view shows only models we track, each at its best.
The newest model with a score here came out in October 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
- 58.0 Gemini-2.5-flash (snapshot unspecified)
- 45.0 GPT-5
- 43.1
DeepSeek-V3.2-Exp
- 42.7 Claude Haiku 4.5
- 39.6 Grok 4 Fast
- 38.6
Llama 3.3 70B
- 37.7
Qwen3-32B
- 37.7 gpt-oss-120b
- 37.3
Gemma 3 27B
- 35.5
Llama 4 Scout
- 35.3
Gemma 3 12B
- 35.1
Mistral Small 3.1
- 33.5
Qwen3-30B-A3B
- 33.1
Aya Expanse 32B
- 30.8 qwen3-4b-instruct-2507
- 30.1
Qwen2.5-3B
- 29.6
Llama 3.1-8B
- 29.1
Aya Expanse 8B
- 28.6
Gemma 3 4B
- 27.0
Llama 3.2 3B
- 1 Gemini-2.5-flash (snapshot unspecified) Zero-shot 58%
- 2 GPT-5 Zero-shot 45%
- 3
DeepSeek-V3.2-Exp Zero-shot 43.1%
- 4 Claude Haiku 4.5 Zero-shot 42.7%
- 5 Grok 4 Fast Zero-shot 39.6%
- 6
Llama 3.3 70B Zero-shot 38.6%
- 7
Qwen3-32B Zero-shot 37.7%
- 8 gpt-oss-120b Zero-shot, low reasoning 37.7%
- 9
Gemma 3 27B Zero-shot 37.3%
- 10
Llama 4 Scout Zero-shot 35.5%
- 11
Gemma 3 12B Zero-shot 35.3%
- 12
Mistral Small 3.1 Zero-shot 35.1%
- 13
Qwen3-30B-A3B Zero-shot, thinking disabled 33.5%
- 14
Aya Expanse 32B Zero-shot 33.1%
- 15 qwen3-4b-instruct-2507 Zero-shot 30.8%
- 16
Qwen2.5-3B Zero-shot 30.1%
- 17
Llama 3.1-8B Zero-shot 29.6%
- 18
Aya Expanse 8B Zero-shot 29.1%
- 19
Gemma 3 4B Zero-shot 28.6%
- 20
Llama 3.2 3B Zero-shot 27%
Source: The IndicParam paper
- 45.0 GPT-5
- 43.1
DeepSeek-V3.2-Exp
- 42.7 Claude Haiku 4.5
- 39.6 Grok 4 Fast
- 38.6
Llama 3.3 70B
- 37.7
Qwen3-32B
- 37.7 gpt-oss-120b
- 37.3
Gemma 3 27B
- 35.5
Llama 4 Scout
- 35.3
Gemma 3 12B
- 35.1
Mistral Small 3.1
- 33.5
Qwen3-30B-A3B
- 33.1
Aya Expanse 32B
- 30.8 qwen3-4b-instruct-2507
- 30.1
Qwen2.5-3B
- 29.6
Llama 3.1-8B
- 29.1
Aya Expanse 8B
- 28.6
Gemma 3 4B
- 27.0
Llama 3.2 3B
- 1 GPT-5 Zero-shot 45%
- 2
DeepSeek-V3.2-Exp Zero-shot 43.1%
- 3 Claude Haiku 4.5 Zero-shot 42.7%
- 4 Grok 4 Fast Zero-shot 39.6%
- 5
Llama 3.3 70B Zero-shot 38.6%
- 6
Qwen3-32B Zero-shot 37.7%
- 7 gpt-oss-120b Zero-shot, low reasoning 37.7%
- 8
Gemma 3 27B Zero-shot 37.3%
- 9
Llama 4 Scout Zero-shot 35.5%
- 10
Gemma 3 12B Zero-shot 35.3%
- 11
Mistral Small 3.1 Zero-shot 35.1%
- 12
Qwen3-30B-A3B Zero-shot, thinking disabled 33.5%
- 13
Aya Expanse 32B Zero-shot 33.1%
- 14 qwen3-4b-instruct-2507 Zero-shot 30.8%
- 15
Qwen2.5-3B Zero-shot 30.1%
- 16
Llama 3.1-8B Zero-shot 29.6%
- 17
Aya Expanse 8B Zero-shot 29.1%
- 18
Gemma 3 4B Zero-shot 28.6%
- 19
Llama 3.2 3B Zero-shot 27%
Source: The IndicParam paper
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Gemini-2.5-flash (snapshot unspecified) | 58% | Zero-shot | Official leaderboard | |
| 2 | GPT-5 | 45% | Zero-shot | Aug 2025 | Official leaderboard |
| 3 | | 43.1% | Zero-shot | Sep 2025 | Official leaderboard |
| 4 | Claude Haiku 4.5 | 42.7% | Zero-shot | Oct 2025 | Official leaderboard |
| 5 | Grok 4 Fast | 39.6% | Zero-shot | Sep 2025 | Official leaderboard |
| 6 | | 38.6% | Zero-shot | Dec 2024 | Official leaderboard |
| 7 | | 37.7% | Zero-shot | Apr 2025 | Official leaderboard |
| 8 | gpt-oss-120b | 37.7% | Zero-shot, low reasoning | Aug 2025 | Official leaderboard |
| 9 | | 37.3% | Zero-shot | Mar 2025 | Official leaderboard |
| 10 | | 35.5% | Zero-shot | Apr 2025 | Official leaderboard |
| 11 | | 35.3% | Zero-shot | Mar 2025 | Official leaderboard |
| 12 | | 35.1% | Zero-shot | Mar 2025 | Official leaderboard |
| 13 | | 33.5% | Zero-shot, thinking disabled | Apr 2025 | Official leaderboard |
| 14 | | 33.1% | Zero-shot | Oct 2024 | Official leaderboard |
| 15 | qwen3-4b-instruct-2507 | 30.8% | Zero-shot | Aug 2025 | Official leaderboard |
| 16 | | 30.1% | Zero-shot | Sep 2024 | Official leaderboard |
| 17 | | 29.6% | Zero-shot | Jul 2024 | Official leaderboard |
| 18 | | 29.1% | Zero-shot | Oct 2024 | Official leaderboard |
| 19 | | 28.6% | Zero-shot | Mar 2025 | Official leaderboard |
| 20 | | 27% | Zero-shot | Sep 2024 | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | GPT-5 | 45% | Zero-shot | Aug 2025 | Official leaderboard |
| 2 | | 43.1% | Zero-shot | Sep 2025 | Official leaderboard |
| 3 | Claude Haiku 4.5 | 42.7% | Zero-shot | Oct 2025 | Official leaderboard |
| 4 | Grok 4 Fast | 39.6% | Zero-shot | Sep 2025 | Official leaderboard |
| 5 | | 38.6% | Zero-shot | Dec 2024 | Official leaderboard |
| 6 | | 37.7% | Zero-shot | Apr 2025 | Official leaderboard |
| 7 | gpt-oss-120b | 37.7% | Zero-shot, low reasoning | Aug 2025 | Official leaderboard |
| 8 | | 37.3% | Zero-shot | Mar 2025 | Official leaderboard |
| 9 | | 35.5% | Zero-shot | Apr 2025 | Official leaderboard |
| 10 | | 35.3% | Zero-shot | Mar 2025 | Official leaderboard |
| 11 | | 35.1% | Zero-shot | Mar 2025 | Official leaderboard |
| 12 | | 33.5% | Zero-shot, thinking disabled | Apr 2025 | Official leaderboard |
| 13 | | 33.1% | Zero-shot | Oct 2024 | Official leaderboard |
| 14 | qwen3-4b-instruct-2507 | 30.8% | Zero-shot | Aug 2025 | Official leaderboard |
| 15 | | 30.1% | Zero-shot | Sep 2024 | Official leaderboard |
| 16 | | 29.6% | Zero-shot | Jul 2024 | Official leaderboard |
| 17 | | 29.1% | Zero-shot | Oct 2024 | Official leaderboard |
| 18 | | 28.6% | Zero-shot | Mar 2025 | Official leaderboard |
| 19 | | 27% | Zero-shot | Sep 2024 | Official leaderboard |
Rankings by part
We read 42 parts from the benchmark’s paper, each ranking models on one part of the test, such as Dogri, Maithili and Rajasthani. GPT-5 leads 27 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.
Dogri
19 models we track have a score here.
- 1 GPT-5 45.7%
- 2 Claude Haiku 4.5 ↑1 44.8%
- 3
Qwen3-32B ↑3 43.4%
- 4
DeepSeek-V3.2-Exp ↓2 43.3%
- 5 Grok 4 Fast ↓1 40.2%
- 6 gpt-oss-120b ↑1 39.7%
- 7
Mistral Small 3.1 ↑4 38%
- 8
Gemma 3 27B 37.6%
- 9
Llama 3.3 70B ↓4 37.3%
- 10
Gemma 3 12B 37.2%
Maithili
19 models we track have a score here.
- 1 GPT-5 43.4%
- 2 Claude Haiku 4.5 ↑1 42.4%
- 3
DeepSeek-V3.2-Exp ↓1 41.4%
- 4 Grok 4 Fast 38.1%
- 5
Llama 3.3 70B 37.5%
- 6
Mistral Small 3.1 ↑5 37.4%
- 7 gpt-oss-120b 35.6%
- 8
Gemma 3 27B 34.4%
- 9
Qwen3-32B ↓3 34%
- 10
Aya Expanse 32B ↑3 33.2%
Rajasthani
19 models we track have a score here.
- 1 GPT-5 44.6%
- 2 Claude Haiku 4.5 ↑1 44.1%
- 3
DeepSeek-V3.2-Exp ↓1 40.1%
- 4
Gemma 3 27B ↑4 39.8%
- 5 gpt-oss-120b ↑2 39.4%
- 6 Grok 4 Fast ↓2 39.1%
- 7
Llama 3.3 70B ↓2 38.2%
- 8
Gemma 3 12B ↑2 35%
- 9
Mistral Small 3.1 ↑2 34.8%
- 10
Llama 4 Scout ↓1 34.8%
Sanskrit
19 models we track have a score here.
- 1 GPT-5 54.8%
- 2
DeepSeek-V3.2-Exp 51.3%
- 3 Claude Haiku 4.5 47.8%
- 4
Llama 4 Scout ↑5 44%
- 5 Grok 4 Fast ↓1 42.9%
- 6
Llama 3.3 70B ↓1 41.7%
- 7 gpt-oss-120b 40.2%
- 8
Mistral Small 3.1 ↑3 40%
- 9
Qwen3-32B ↓3 39.2%
- 10
Gemma 3 27B ↓2 38.6%
Sans-Eng
19 models we track have a score here.
- 1 GPT-5 64.6%
- 2
DeepSeek-V3.2-Exp 61.4%
- 3 Claude Haiku 4.5 58.1%
- 4 Grok 4 Fast 54.2%
- 5
Llama 3.3 70B 51.6%
- 6
Gemma 3 27B ↑2 49.1%
- 7
Llama 4 Scout ↑2 48.7%
- 8
Qwen3-32B ↓2 48.6%
- 9 gpt-oss-120b ↓2 48.1%
- 10
Qwen3-30B-A3B ↑2 44.2%
Bodo
19 models we track have a score here.
- 1
DeepSeek-V3.2-Exp ↑1 36.4%
- 2
Qwen3-32B ↑4 31.8%
- 3 GPT-5 ↓2 31.6%
- 4
Gemma 3 12B ↑6 30.6%
- 5 Grok 4 Fast ↓1 30.6%
- 6
Mistral Small 3.1 ↑5 30.1%
- 7
Qwen3-30B-A3B ↑5 29.9%
- 8
Gemma 3 27B 28.9%
- 9
Llama 3.3 70B ↓4 28.6%
- 10
Qwen2.5-3B ↑5 28.3%
Santali
19 models we track have a score here.
- 1
DeepSeek-V3.2-Exp ↑1 41.5%
- 2 GPT-5 ↓1 40.4%
- 3 Claude Haiku 4.5 40.4%
- 4
Llama 3.3 70B ↑1 40.1%
- 5
Qwen3-32B ↑1 39.7%
- 6 Grok 4 Fast ↓2 38.8%
- 7
Gemma 3 27B ↑1 36.7%
- 8
Llama 4 Scout ↑1 36.3%
- 9
Qwen3-30B-A3B ↑3 35.9%
- 10
Mistral Small 3.1 ↑1 35.7%
Konkani
19 models we track have a score here.
- 1 Claude Haiku 4.5 ↑2 42.1%
- 2 GPT-5 ↓1 39%
- 3
DeepSeek-V3.2-Exp ↓1 37.6%
- 4 Grok 4 Fast 37.4%
- 5
Llama 3.3 70B 37.3%
- 6
Gemma 3 27B ↑2 36.5%
- 7
Gemma 3 12B ↑3 35.7%
- 8 gpt-oss-120b ↓1 35.4%
- 9
Llama 4 Scout 34.6%
- 10
Qwen3-32B ↓4 34.5%
Extremely low resource average
19 models we track have a score here.
- 1 GPT-5 45.1%
- 2
DeepSeek-V3.2-Exp 43.7%
- 3 Claude Haiku 4.5 43%
- 4 Grok 4 Fast 39.7%
- 5
Qwen3-30B-A3B ↑7 38.6%
- 6
Llama 3.3 70B ↓1 37.6%
- 7 gpt-oss-120b 37.4%
- 8
Gemma 3 27B 37.3%
- 9
Mistral Small 3.1 ↑2 36.2%
- 10
Llama 4 Scout ↓1 35.4%
Nepali
19 models we track have a score here.
- 1 Claude Haiku 4.5 ↑2 43.8%
- 2 GPT-5 ↓1 43.4%
- 3 Grok 4 Fast ↑1 40.9%
- 4
DeepSeek-V3.2-Exp ↓2 39.7%
- 5
Gemma 3 27B ↑3 39.3%
- 6
Qwen3-30B-A3B ↑6 37.2%
- 7 gpt-oss-120b 37.1%
- 8
Llama 3.3 70B ↓3 35%
- 9
Mistral Small 3.1 ↑2 34.9%
- 10
Gemma 3 12B 34.8%
Gujarati
19 models we track have a score here.
- 1 GPT-5 40%
- 2
DeepSeek-V3.2-Exp 39.5%
- 3 gpt-oss-120b ↑4 37%
- 4 Claude Haiku 4.5 ↓1 37%
- 5
Llama 3.3 70B 35.9%
- 6 Grok 4 Fast ↓2 35.9%
- 7
Gemma 3 12B ↑3 33.8%
- 8
Qwen3-30B-A3B ↑4 33.5%
- 9
Mistral Small 3.1 ↑2 32.6%
- 10
Gemma 3 27B ↓2 31.1%
Marathi
19 models we track have a score here.
- 1 GPT-5 48.4%
- 2
DeepSeek-V3.2-Exp 44.6%
- 3 Claude Haiku 4.5 43.9%
- 4
Llama 4 Scout ↑5 43.5%
- 5
Qwen3-30B-A3B ↑7 43.5%
- 6
Llama 3.3 70B ↓1 40.8%
- 7
Gemma 3 27B ↑1 39.9%
- 8 Grok 4 Fast ↓4 39.8%
- 9 gpt-oss-120b ↓2 39.7%
- 10
Gemma 3 12B 38.4%
Odia
19 models we track have a score here.
- 1 GPT-5 48.7%
- 2 Claude Haiku 4.5 ↑1 44%
- 3
DeepSeek-V3.2-Exp ↓1 42.1%
- 4
Llama 3.3 70B ↑1 40.7%
- 5 Grok 4 Fast ↓1 40.7%
- 6
Qwen3-30B-A3B ↑6 40.4%
- 7
Gemma 3 27B ↑1 39.9%
- 8 gpt-oss-120b ↓1 39.9%
- 9
Llama 4 Scout 35%
- 10 qwen3-4b-instruct-2507 ↑4 34.7%
Low resource average
19 models we track have a score here.
- 1 GPT-5 44.9%
- 2 Claude Haiku 4.5 ↑1 42%
- 3
DeepSeek-V3.2-Exp ↓1 41.6%
- 4 Grok 4 Fast 39.2%
- 5
Qwen3-30B-A3B ↑7 38.7%
- 6 gpt-oss-120b ↑1 38.3%
- 7
Llama 3.3 70B ↓2 37.9%
- 8
Gemma 3 27B 37.4%
- 9
Gemma 3 12B ↑1 35.6%
- 10
Llama 4 Scout ↓1 35.5%
MCQ
19 models we track have a score here.
- 1 GPT-5 46.1%
- 2 Claude Haiku 4.5 ↑1 43.9%
- 3
DeepSeek-V3.2-Exp ↓1 43.2%
- 4 Grok 4 Fast 40.4%
- 5
Qwen3-30B-A3B ↑7 39.9%
- 6
Gemma 3 27B ↑2 38.1%
- 7 gpt-oss-120b 37.9%
- 8
Llama 4 Scout ↑1 37.5%
- 9
Llama 3.3 70B ↓4 37%
- 10
Mistral Small 3.1 ↑1 35.7%
List Matching
19 models we track have a score here.
- 1
DeepSeek-V3.2-Exp ↑1 47.2%
- 2 GPT-5 ↓1 41.3%
- 3
Llama 3.3 70B ↑2 39.7%
- 4 Grok 4 Fast 39.2%
- 5 Claude Haiku 4.5 ↓2 39%
- 6 gpt-oss-120b ↑1 38%
- 7
Aya Expanse 32B ↑6 36.4%
- 8
Gemma 3 12B ↑2 35.9%
- 9
Mistral Small 3.1 ↑2 34.3%
- 10
Qwen3-30B-A3B ↑2 33.1%
A&R
19 models we track have a score here.
- 1
DeepSeek-V3.2-Exp ↑1 44.5%
- 2
Llama 3.3 70B ↑3 41.6%
- 3 GPT-5 ↓2 41.4%
- 4 Claude Haiku 4.5 ↓1 40%
- 5 gpt-oss-120b ↑2 38.7%
- 6
Qwen3-30B-A3B ↑6 37.9%
- 7
Gemma 3 27B ↑1 37.2%
- 8
Gemma 3 12B ↑2 36.7%
- 9
Llama 4 Scout 34.6%
- 10 Grok 4 Fast ↓6 33.7%
Blanks
19 models we track have a score here.
- 1 GPT-5 47.8%
- 2
DeepSeek-V3.2-Exp 45.9%
- 3
Qwen3-30B-A3B ↑9 40.1%
- 4
Gemma 3 27B ↑4 38.9%
- 5 gpt-oss-120b ↑2 38.2%
- 6 Claude Haiku 4.5 ↓3 37.6%
- 7
Aya Expanse 32B ↑6 36.9%
- 8 Grok 4 Fast ↓4 36.9%
- 9
Llama 3.1-8B ↑7 35%
- 10
Llama 4 Scout ↓1 35%
IS
19 models we track have a score here.
- 1 Claude Haiku 4.5 ↑2 45.7%
- 2 GPT-5 ↓1 43.3%
- 3
Llama 3.3 70B ↑2 40.7%
- 4 Grok 4 Fast 38.5%
- 5
DeepSeek-V3.2-Exp ↓3 38%
- 6
Gemma 3 12B ↑4 37.1%
- 7
Gemma 3 27B ↑1 36.3%
- 8
Mistral Small 3.1 ↑3 35.8%
- 9
Qwen3-30B-A3B ↑3 35.6%
- 10 gpt-oss-120b ↓3 32.5%
Ordering
19 models we track have a score here.
- 1 GPT-5 41.6%
- 2
Llama 3.3 70B ↑3 38.1%
- 3 gpt-oss-120b ↑4 37.4%
- 4
DeepSeek-V3.2-Exp ↓2 37.1%
- 5
Mistral Small 3.1 ↑6 37%
- 6 Grok 4 Fast ↓2 36.9%
- 7 Claude Haiku 4.5 ↓4 36.6%
- 8
Gemma 3 27B 35.7%
- 9
Gemma 3 12B ↑1 34.8%
- 10
Qwen3-30B-A3B ↑2 34%
Knowledge: Gujarati
19 models we track have a score here.
- 1 GPT-5 40.1%
- 2
DeepSeek-V3.2-Exp 39.6%
- 3 gpt-oss-120b ↑4 37%
- 4 Claude Haiku 4.5 ↓1 37%
- 5
Qwen3-32B ↑1 35.9%
- 6 Grok 4 Fast ↓2 35.9%
- 7
Gemma 3 12B ↑3 33.9%
- 8
Llama 3.3 70B ↓3 33.6%
- 9
Mistral Small 3.1 ↑2 32.3%
- 10
Gemma 3 27B ↓2 31.3%
Knowledge: Konkani
19 models we track have a score here.
- 1 Claude Haiku 4.5 ↑2 41.9%
- 2 GPT-5 ↓1 38.5%
- 3
DeepSeek-V3.2-Exp ↓1 37.3%
- 4
Llama 3.3 70B ↑1 37.2%
- 5 Grok 4 Fast ↓1 37.1%
- 6
Gemma 3 27B ↑2 36%
- 7
Gemma 3 12B ↑3 35.5%
- 8 gpt-oss-120b ↓1 34.9%
- 9
Llama 4 Scout 34.4%
- 10
Qwen3-32B ↓4 34.2%
Knowledge: Maithili
19 models we track have a score here.
- 1 GPT-5 41.4%
- 2 Claude Haiku 4.5 ↑1 40.3%
- 3
DeepSeek-V3.2-Exp ↓1 39.7%
- 4 Grok 4 Fast 36.5%
- 5
Mistral Small 3.1 ↑6 36.3%
- 6
Llama 3.3 70B ↓1 35.2%
- 7 gpt-oss-120b 33.9%
- 8
Gemma 3 27B 33.2%
- 9
Aya Expanse 32B ↑4 31.7%
- 10
Qwen3-32B ↓4 31.5%
Knowledge: Marathi
19 models we track have a score here.
- 1 GPT-5 47.7%
- 2
DeepSeek-V3.2-Exp 43.9%
- 3 Claude Haiku 4.5 42.9%
- 4
Llama 3.3 70B ↑1 42.7%
- 5
Llama 4 Scout ↑4 42.1%
- 6
Qwen3-32B 40%
- 7
Gemma 3 27B ↑1 38.9%
- 8 Grok 4 Fast ↓4 38.8%
- 9 gpt-oss-120b ↓2 38.7%
- 10
Gemma 3 12B 37.5%
Knowledge: Oriya
19 models we track have a score here.
- 1 GPT-5 44.1%
- 2 Claude Haiku 4.5 ↑1 39.8%
- 3
Llama 3.3 70B ↑2 39.3%
- 4
DeepSeek-V3.2-Exp ↓2 38.7%
- 5
Qwen3-32B ↑1 38.5%
- 6 Grok 4 Fast ↓2 38.3%
- 7
Gemma 3 27B ↑1 37.8%
- 8 gpt-oss-120b ↓1 36.3%
- 9
Qwen3-30B-A3B ↑3 33.9%
- 10
Gemma 3 12B 33%
Knowledge: Rajasthani
19 models we track have a score here.
- 1 GPT-5 40.8%
- 2 Claude Haiku 4.5 ↑1 39.7%
- 3
DeepSeek-V3.2-Exp ↓1 37.4%
- 4 Grok 4 Fast 37.1%
- 5
Gemma 3 27B ↑3 36.9%
- 6
Llama 3.3 70B ↓1 36.6%
- 7 gpt-oss-120b 35.5%
- 8
Aya Expanse 32B ↑5 33%
- 9
Gemma 3 12B ↑1 32.6%
- 10
Mistral Small 3.1 ↑1 32.3%
Knowledge: Sanskrit
19 models we track have a score here.
- 1 GPT-5 57.8%
- 2
DeepSeek-V3.2-Exp 53.5%
- 3 Claude Haiku 4.5 48.8%
- 4 Grok 4 Fast 45.2%
- 5
Llama 4 Scout ↑4 44.3%
- 6
Llama 3.3 70B ↓1 42.8%
- 7 gpt-oss-120b 41.5%
- 8
Mistral Small 3.1 ↑3 41.3%
- 9
Qwen3-32B ↓3 40.1%
- 10
Gemma 3 27B ↓2 39.9%
Knowledge: Sans-Eng
19 models we track have a score here.
- 1 GPT-5 66.3%
- 2
DeepSeek-V3.2-Exp 62.9%
- 3 Claude Haiku 4.5 59.5%
- 4 Grok 4 Fast 55.9%
- 5
Llama 3.3 70B 52.1%
- 6 gpt-oss-120b ↑1 50.2%
- 7
Qwen3-32B ↓1 49.5%
- 8
Gemma 3 27B 49.4%
- 9
Llama 4 Scout 49.4%
- 10
Mistral Small 3.1 ↑1 45.2%
Knowledge: Dogri
19 models we track have a score here.
- 1 GPT-5 45.1%
- 2 Claude Haiku 4.5 ↑1 44.9%
- 3
Qwen3-32B ↑3 43.1%
- 4
DeepSeek-V3.2-Exp ↓2 42.8%
- 5 Grok 4 Fast ↓1 39.9%
- 6 gpt-oss-120b ↑1 38.7%
- 7
Mistral Small 3.1 ↑4 38%
- 8
Gemma 3 27B 37.8%
- 9
Llama 3.3 70B ↓4 37.5%
- 10
Gemma 3 12B 36.7%
Knowledge: Nepali
19 models we track have a score here.
- 1 Claude Haiku 4.5 ↑2 43.4%
- 2 GPT-5 ↓1 41.5%
- 3 Grok 4 Fast ↑1 40.6%
- 4
DeepSeek-V3.2-Exp ↓2 39.5%
- 5
Gemma 3 27B ↑3 39.4%
- 6 gpt-oss-120b ↑1 36.1%
- 7
Llama 3.3 70B ↓2 35.8%
- 8
Gemma 3 12B ↑2 34.8%
- 9
Mistral Small 3.1 ↑2 34.8%
- 10
Llama 4 Scout ↓1 34%
Knowledge: Santali
19 models we track have a score here.
- 1
DeepSeek-V3.2-Exp ↑1 40.7%
- 2 Claude Haiku 4.5 ↑1 40.3%
- 3 GPT-5 ↓2 40.1%
- 4
Qwen3-32B ↑2 39.7%
- 5
Llama 3.3 70B 39%
- 6 Grok 4 Fast ↓2 38.1%
- 7
Gemma 3 27B ↑1 36.7%
- 8
Mistral Small 3.1 ↑3 35.8%
- 9
Llama 4 Scout 35.7%
- 10 gpt-oss-120b ↓3 35.7%
Language understanding: Gujarati
19 models we track have a score here.
- 1
Mistral Small 3.1 ↑10 83.3%
- 2
Qwen2.5-3B ↑13 66.7%
- 3
Aya Expanse 8B ↑14 50%
- 4
Qwen3-30B-A3B ↑8 50%
- 5 qwen3-4b-instruct-2507 ↑9 50%
- 6
Aya Expanse 32B ↑7 33.3%
- 7
Qwen3-32B ↓1 33.3%
- 8 gpt-oss-120b ↓1 33.3%
- 9 GPT-5 ↓8 33.3%
- 10 Grok 4 Fast ↓6 33.3%
Language understanding: Konkani
19 models we track have a score here.
- 1 GPT-5 60.6%
- 2
Aya Expanse 32B ↑11 57.6%
- 3
Gemma 3 27B ↑5 57.6%
- 4 gpt-oss-120b ↑3 54.5%
- 5
Llama 3.2 3B ↑14 51.5%
- 6 Grok 4 Fast ↓2 51.5%
- 7 Claude Haiku 4.5 ↓4 51.5%
- 8
DeepSeek-V3.2-Exp ↓6 48.5%
- 9
Llama 3.1-8B ↑7 45.5%
- 10
Mistral Small 3.1 ↑1 45.5%
Language understanding: Maithili
19 models we track have a score here.
- 1 GPT-5 60.8%
- 2 Claude Haiku 4.5 ↑1 60.8%
- 3
Llama 3.3 70B ↑2 57.7%
- 4
DeepSeek-V3.2-Exp ↓2 56.9%
- 5
Qwen3-32B ↑1 56.2%
- 6 Grok 4 Fast ↓2 52.3%
- 7 gpt-oss-120b 50.8%
- 8
Gemma 3 12B ↑2 49.2%
- 9
Mistral Small 3.1 ↑2 46.9%
- 10
Aya Expanse 32B ↑3 46.2%
Language understanding: Marathi
19 models we track have a score here.
- 1
Llama 4 Scout ↑8 70.7%
- 2 GPT-5 ↓1 63.8%
- 3 Claude Haiku 4.5 63.8%
- 4
Gemma 3 27B ↑4 60.3%
- 5 gpt-oss-120b ↑2 60.3%
- 6 Grok 4 Fast ↓2 60.3%
- 7
Llama 3.3 70B ↓2 58.6%
- 8
DeepSeek-V3.2-Exp ↓6 58.6%
- 9
Gemma 3 12B ↑1 56.9%
- 10
Qwen3-32B ↓4 56.9%
Language understanding: Oriya
19 models we track have a score here.
- 1 GPT-5 66.7%
- 2 Claude Haiku 4.5 ↑1 60.7%
- 3
DeepSeek-V3.2-Exp ↓1 55.6%
- 4 gpt-oss-120b ↑3 53.8%
- 5 Grok 4 Fast ↓1 50.4%
- 6
Qwen3-32B 49.6%
- 7
Gemma 3 27B ↑1 47.9%
- 8
Llama 4 Scout ↑1 46.2%
- 9
Llama 3.3 70B ↓4 44.4%
- 10 qwen3-4b-instruct-2507 ↑4 42.7%
Language understanding: Rajasthani
19 models we track have a score here.
- 1 Claude Haiku 4.5 ↑2 55.8%
- 2 GPT-5 ↓1 54.5%
- 3 gpt-oss-120b ↑4 49.7%
- 4
Gemma 3 27B ↑4 47.6%
- 5
DeepSeek-V3.2-Exp ↓3 47%
- 6
Qwen3-32B 44.8%
- 7 Grok 4 Fast ↓3 44.2%
- 8
Llama 3.3 70B ↓3 42.4%
- 9
Gemma 3 12B ↑1 41.5%
- 10
Llama 4 Scout ↓1 41.5%
Language understanding: Sanskrit
19 models we track have a score here.
- 1 Claude Haiku 4.5 ↑2 43.6%
- 2 GPT-5 ↓1 43.2%
- 3
Llama 4 Scout ↑6 42.8%
- 4
DeepSeek-V3.2-Exp ↓2 42.8%
- 5
Llama 3.3 70B 37.5%
- 6
Qwen3-32B 36%
- 7
Mistral Small 3.1 ↑4 34.8%
- 8 gpt-oss-120b ↓1 34.8%
- 9 Grok 4 Fast ↓5 33.7%
- 10
Gemma 3 27B ↓2 33.3%
Language understanding: Sans-Eng
19 models we track have a score here.
- 1 GPT-5 51.4%
- 2
DeepSeek-V3.2-Exp 49.5%
- 3
Llama 3.3 70B ↑2 47.7%
- 4
Gemma 3 27B ↑4 46.8%
- 5 Claude Haiku 4.5 ↓2 46.8%
- 6
Llama 4 Scout ↑3 43.2%
- 7
Qwen3-32B ↓1 41.4%
- 8
Qwen3-30B-A3B ↑4 40.5%
- 9 Grok 4 Fast ↓5 40.5%
- 10
Mistral Small 3.1 ↑1 35.1%
Language understanding: Dogri
19 models we track have a score here.
- 1 GPT-5 48.4%
- 2
DeepSeek-V3.2-Exp 45.7%
- 3
Qwen3-32B ↑3 44.7%
- 4 gpt-oss-120b ↑3 44.1%
- 5 Claude Haiku 4.5 ↓2 44.1%
- 6 Grok 4 Fast ↓2 41.5%
- 7
Gemma 3 12B ↑3 39.4%
- 8
Mistral Small 3.1 ↑3 37.8%
- 9
Gemma 3 27B ↓1 36.7%
- 10
Llama 3.3 70B ↓5 36.2%
Language understanding: Nepali
19 models we track have a score here.
- 1 GPT-5 51.8%
- 2 Claude Haiku 4.5 ↑1 45.6%
- 3
Llama 3.3 70B ↑2 43.1%
- 4 Grok 4 Fast 42.6%
- 5 gpt-oss-120b ↑2 41.5%
- 6
Qwen3-32B 41%
- 7
DeepSeek-V3.2-Exp ↓5 40.5%
- 8
Gemma 3 27B 39%
- 9
Llama 4 Scout 36.9%
- 10
Aya Expanse 32B ↑3 35.4%
Language understanding: Santali
19 models we track have a score here.
- 1
Llama 3.3 70B ↑4 48.5%
- 2
DeepSeek-V3.2-Exp 47.5%
- 3 Grok 4 Fast ↑1 44.4%
- 4 GPT-5 ↓3 43.4%
- 5
Llama 4 Scout ↑4 41.4%
- 6 Claude Haiku 4.5 ↓3 41.4%
- 7
Qwen3-30B-A3B ↑5 40.4%
- 8
Qwen3-32B ↓2 40.4%
- 9
Aya Expanse 32B ↑4 36.4%
- 10
Aya Expanse 8B ↑7 36.4%
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: The IndicParam 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 IndicParam 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 Multilingual
- MMMLU In use MMLU professionally translated into 14 languages. Multiple-choice accuracy. Claude Mythos Preview 92.7%
- WMT24++ In use English-to-55-language translation of WMT24 test data. Scored with automatic metrics such as MetricX or xCOMET.
K-EXAONE 90.5%
- MMLU-ProX In use MMLU-Pro translated into up to 29 languages with expert review. Multiple-choice accuracy per language.
Qwen3.7-Max 87%
- CLIcK In use Models answer 1,995 multiple-choice questions on Korean language and culture. Questions come from official Korean exams and textbooks in eleven categories.
Solar Open 2 90.7%
- Global MMLU Lite 3.0 In use Models answer 400 human-translated MMLU questions per language in 23 languages. Version 3.0 adds Czech, Hungarian, Oriya, Slovak and Tajik and updates Italian. Accuracy. Claude Opus 5 95.7%
- MMMB In use Models answer multiple-choice questions about images in six languages across 15 categories. The set has about 12,000 questions. LFM2.5-VL-3B 83%
Data audit
9 of 10 checks pass, 1 warning. Look first at the Newest score check. The newest came 13 months ago, in October 2025, for Claude Haiku 4.5. 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 (20 entries).
- Pass Models scored 19 models we track, plus 1 other entry on its leaderboard.
- Warning Newest score The newest came 13 months ago, in October 2025, for Claude Haiku 4.5.
- Note Frontier models None of the top 10 has a score. The strongest tested is GPT-5 (#71).
- Pass Headroom The best is 45%, under 85%, so it still separates models.
Leaderboard import 3 pass
- Pass Importer We read the board on aclanthology.org: 19 rows for models we track and 1 other entry 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.