Benchmark Almanac

AraTrust

AbdulmalekDS qwen72b-ar-lora 91.4% tops its leaderboard Qwen2-72B 89.2% leads the models we track Dropped

Models answer 522 human-written Arabic multiple-choice questions on truthfulness, ethics, safety, physical and mental health, unfairness, illegal activities, privacy and offensive language.

It is a task in the Open Arabic LLM Leaderboard v2.

In the authors’ words

The swift progress and widespread acceptance of artificial intelligence (AI) systems highlight a pressing requirement to comprehend both the capabilities and potential risks associated with AI. Given the linguistic complexity, cultural richness, and underrepresented status of Arabic in AI research, there is a pressing need to focus on Large Language Models (LLMs) performance and safety for Arabic-related tasks. Despite some progress in their development, there is a lack of comprehensive trustworthiness evaluation benchmarks, which presents a major challenge in accurately assessing and improving the safety of LLMs when prompted in Arabic. In this paper, we introduce AraTrust, the first comprehensive trustworthiness benchmark for LLMs in Arabic. AraTrust comprises 522 human-written multiple-choice questions addressing diverse dimensions related to truthfulness, ethics, safety, physical health, mental health, unfairness, illegal activities, privacy, and offensive language. We evaluated a set of LLMs against our benchmark to assess their trustworthiness. GPT-4 was the most trustworthy LLM, while open-source models, particularly AceGPT 7B and Jais 13B, struggled to achieve a score of 60% in our benchmark.

Abstract of AraTrust: An Evaluation of Trustworthiness for LLMs in Arabic (arXiv:2403.09017, CC0)
Made by Alghamdi et al. Released Mar 2024 Deception and alignment All tasks public Last in a launch post May 2025 , dropped after it saturated Last model added Nov 2025 Scores refreshed 2 Nov 2026
Level
Everyday Most adults, without training
Scored
Checked automatically A program checks the result against an answer key, unit tests or the expected final state.
Answers
Multiple choice
Paper Leaderboard
150
Entries on its own leaderboard, 39 models we track
1
Launch post lists it, from 1 lab
1
System card reports it, from 1 lab

AbdulmalekDS qwen72b-ar-lora tops the leaderboard with 91.4%

Best score per entry from the benchmark's own leaderboard, read on 2 Nov 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. The best model we track is Qwen2-72B, at 89.2%.

The newest model with a score here came out in Aug 2025, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

150 entries. Scroll sideways for the rest.

  1. 91.4 AbdulmalekDS qwen72b-ar-lora
  2. 91.0 Lina-Z arabic_llm_test
  3. 90.6 Mushari440 Qwen3-8B-SFT-V2
  4. 90.0 Cran-May tempmotacilla-cinerea-0308
  5. 90.0 hammh0a Hala-9B
  6. 89.8 rombodawg Rombos-LLM-V2.5-Qwen-72b
  7. 89.8 Applied-Innovation-Center AIC-1
  8. 89.7 deep-analysis-research D2IL-Arabic-Qwen2.5-72B-Instruct-v0.1
  9. 89.7 tunny Arabic_Qwen2.5_72B_instruct_finetune_0.1
  10. 89.7 Lina-Z qwen_arabic_ft
  11. 89.7 Sakalti Ultiima-72B
  12. 89.5 deep-analysis-research D2IL-Arabic-Qwen2.5-72B-Instruct-v0.2
  13. 89.2 Qwen2-72B
  14. 89.2 MaziyarPanahi calme-2.2-qwen2.5-72b
  15. 89.2 Qwen2.5-72B
  16. 89.2 AXCXEPT EZO-Qwen2.5-32B-Instruct
  17. 89.1 Gemma 3 27B
  18. 89.1 MaziyarPanahi calme-2.1-qwen2.5-72b
  19. 89.0 Aya Expanse 32B
  20. 88.8 tanliboy lambda-qwen2.5-14b-dpo-test
  21. 88.8 Sakalti SakaMoe-3x14B-Instruct
  22. 88.4 zetasepic Qwen2.5-32B-Instruct-abliterated-v2
  23. 88.3 Applied-Innovation-Center Karnak
  24. 88.3 Sakalti Saka-14B
  25. 88.3 QCRI/Fanar-1-9B-Instruct
  26. 88.3 Llama 3.3 70B Instruct
  27. 88.3 maldv Awqward2.5-32B-Instruct
  28. 88.1 maldv Qwentile2.5-32B-Instruct
  29. 87.8 TIGER-Lab Qwen2.5-32B-Instruct-CFT
  30. 87.7 Daemontatox Cogito-R1
  31. 87.6 TarjamaN Pronoia-14b-community
  32. 87.5 Navid-AI Yehia-7B-preview
  33. 87.3 Mushari440 qwen3-8B-SFT
  34. 87.0 huihui-ai Qwen2.5-32B-Instruct-abliterated
  35. 87.0 v000000 Qwen2.5-Lumen-14B
  36. 87.0 Gemma 3 12B
  37. 86.9 ALLaM-AI/ALLaM-7B-Instruct-preview
  38. 86.9 v000000 Qwen2.5-14B-Gutenberg-1e-Delta
  39. 86.8 freewheelin free-evo-qwen72b-v0.8-re
  40. 86.8 rombodawg Rombos-LLM-V2.6-Qwen-14b
  41. 86.7 FreedomIntelligence AceGPT-v2-70B-Chat
  42. 86.7 Qwen2.5-32B
  43. 86.5 recoilme recoilme-gemma-2-9B-v0.2
  44. 86.5 tanliboy lambda-qwen2.5-32b-dpo-test
  45. 86.5 Qwen2.5-14B
  46. 86.2 Slim205 Barka-9b-it
  47. 85.5 Mushari440 qwen-3-8B-sft-grpo
  48. 84.9 Qwen3-4B
  49. 84.7 FreedomIntelligence AceGPT-v2-32B-Chat
  50. 84.5 recoilme recoilme-gemma-2-9B-v0.4
  51. 84.3 FreedomIntelligence AceGPT-v2-70B
  52. 84.3 Ocelotr Qwen3-8B-GAE
  53. 84.2 CohereForAI/aya-23-35B
  54. 84.0 amine-khelif BARS
  55. 84.0 asas-ai AceGPT-v2-32B
  56. 84.0 FreedomIntelligence AceGPT-v2-32B
  57. 83.7 inceptionai/jais-adapted-70b-chat
  58. 83.4 Qwen3-8B
  59. 83.1 inceptionai/jais-family-30b-8k-chat
  60. 82.8 silma-ai SILMA-9B-Instruct-v1.0
  61. 82.8 Qwen2-7B-Instruct
  62. 82.6 arcee-ai Arcee-Spark
  63. 82.5 Aya Expanse 8B
  64. 82.1 QCRI/Fanar-1-9B
  65. 82.1 Cran-May T.E-8.1
  66. 81.9 JasperV13 Yehia-7B-DPO-Reasoning-preview
  67. 81.8 airev-ai emirati-14b-v2
  68. 81.6 inceptionai/jais-family-30b-16k-chat
  69. 81.3 huihui-ai Qwen2.5-7B-Instruct-abliterated
  70. 80.8 SeaLLMs SeaLLM-7B-v2.5
  71. 80.8 JasperV13 Yehia-7B-Reasoning-preview
  72. 80.7 Qwen2.5-7B
  73. 80.5 Gemma 2 27B
  74. 80.5 CohereForAI/c4ai-command-r7b-arabic-02-2025
  75. 80.4 Orion-zhen Qwen2.5-7B-Instruct-Uncensored
  76. 79.7 inceptionai/jais-adapted-13b-chat
  77. 78.3 Gemma 3 4B
  78. 78.1 Qwen1.5-32B
  79. 78.0 SeaLLMs SeaLLMs-v3-7B-Chat
  80. 77.8 Phi-4 Mini
  81. 77.2 Gemma 2 9B
  82. 76.6 JasperV13 Yehia-7B-SFT-Reasoning-preview
  83. 76.5 FreedomIntelligence AceGPT-v1.5-13B-Chat
  84. 76.5 Sakalti oxyge1-33B
  85. 75.6 Llama 3.1-8B
  86. 75.6 Qwen3-1.7B
  87. 75.0 inceptionai/jais-family-13b-chat
  88. 75.0 inceptionai/jais-adapted-70b
  89. 74.4 AIDC-AI Marco-LLM-AR-V4
  90. 74.2 huihui-ai Qwen2.5-7B-Instruct-abliterated-v2
  91. 74.2 FreedomIntelligence AceGPT-v2-8B-Chat
  92. 73.5 AIDC-AI Marco-LLM-AR-V2
  93. 73.3 silma-ai SILMA-Kashif-2B-Instruct-v1.0
  94. 73.3 Ocelotr Qwen-8B-SFT
  95. 72.7 Qwen3-30B-A3B-Instruct
  96. 72.5 Qwen2.5-3B
  97. 72.2 Isaak-Carter Josiefied-Qwen2.5-7B-Instruct-abliterated-v2
  98. 71.5 Slim205 Barka-2b-it
  99. 70.8 Qwen2.5-1.5B
  100. 70.7 inceptionai/jais-family-30b-8k
  101. 69.8 Hagrass LLama3-3.2-instruct-trained
  102. 68.9 Qwen2-1.5B-Instruct
  103. 68.6 AIDC-AI Marco-LLM-AR
  104. 67.2 Qwen1.5-7B
  105. 66.0 inceptionai/jais-adapted-7b-chat
  106. 65.7 FreedomIntelligence AceGPT-13B-chat
  107. 65.5 upstage SOLAR-10.7B-v1.0
  108. 64.3 Qwen1.5-14B
  109. 64.3 VAGOsolutions SauerkrautLM-Nemo-12b-Instruct
  110. 63.6 FreedomIntelligence AceGPT-7B-chat
  111. 62.0 inceptionai/jais-family-30b-16k
  112. 61.2 inceptionai/jais-adapted-13b
  113. 61.0 01-ZeroOne SUHAIL-14B-preview
  114. 58.5 Llama 3.2 3B
  115. 58.1 01-ai/Yi-1.5-9B
  116. 58.0 tiiuae/Falcon3-10B-Base
  117. 57.5 Qwen3-14B
  118. 57.2 Qwen/Qwen1.5-4B
  119. 56.6 01-ai/Yi-1.5-9B-32K
  120. 56.5 FreedomIntelligence AceGPT-13B
  121. 56.5 Naseej noon-7b
  122. 55.8 Syed-Hasan-8503 Phi-3-mini-4K-instruct-cpo-simpo
  123. 55.4 speakleash Bielik-11B-v2
  124. 54.2 cognitivecomputations Dolphin3.0-R1-Mistral-24B
  125. 52.3 01-ai/Yi-1.5-34B-32K
  126. 51.5 Qwen3-0.6B
  127. 49.9 tiiuae/Falcon3-7B-Base
  128. 49.4 FreedomIntelligence AceGPT-7B
  129. 47.7 FreedomIntelligence AceGPT-v1.5-13B
  130. 47.6 Phi-4
  131. 47.5 Sakalti Saka-1.5B
  132. 47.3 Qwen2.5-0.5B
  133. 46.9 Qwen2-0.5B-Instruct
  134. 46.7 INSAIT-Institute BgGPT-7B-Instruct-v0.2
  135. 40.6 qwen3-4b-instruct-2507
  136. 39.6 AIDC-AI Marco-LLM-AR-V3
  137. 38.9 Qwen3-32B
  138. 37.1 Llama 3.2 1B
  139. 34.8 inceptionai/jais-family-13b
  140. 33.9 inceptionai/jais-adapted-7b
  141. 32.8 riotu-lab ArabianGPT-01B
  142. 32.6 mobiuslabsgmbh DeepSeek-R1-ReDistill-Llama3-8B-v1.1
  143. 32.6 mobiuslabsgmbh DeepSeek-R1-ReDistill-Qwen-7B-v1.1
  144. 31.5 Qwen/Qwen1.5-1.8B
  145. 31.2 Gemma 2 2B IT
  146. 31.2 DeepSeek-R1-Distill-Qwen-14B
  147. 31.2 DeepSeek-R1-Distill-Qwen-32B
  148. 31.2 QwQ-32B
  149. 31.2 stabilityai ar-stablelm-2-base
  150. 30.9 TIGER-Lab Qwen2.5-Math-7B-CFT
  1. 1 AbdulmalekDS qwen72b-ar-lora 91.4%
  2. 2 Lina-Z arabic_llm_test 91%
  3. 3 Mushari440 Qwen3-8B-SFT-V2 90.6%
  4. 4 Cran-May tempmotacilla-cinerea-0308 90%
  5. 5 hammh0a Hala-9B 90%
  6. 6 rombodawg Rombos-LLM-V2.5-Qwen-72b 89.8%
  7. 7 Applied-Innovation-Center AIC-1 89.8%
  8. 8 deep-analysis-research D2IL-Arabic-Qwen2.5-72B-Instruct-v0.1 89.7%
  9. 9 tunny Arabic_Qwen2.5_72B_instruct_finetune_0.1 89.7%
  10. 10 Lina-Z qwen_arabic_ft 89.7%
  11. 11 Sakalti Ultiima-72B 89.7%
  12. 12 deep-analysis-research D2IL-Arabic-Qwen2.5-72B-Instruct-v0.2 89.5%
  13. 13 Qwen2-72B base model 89.2%
  14. 14 MaziyarPanahi calme-2.2-qwen2.5-72b 89.2%
  15. 15 Qwen2.5-72B 89.2%
  16. 16 AXCXEPT EZO-Qwen2.5-32B-Instruct 89.2%
  17. 17 Gemma 3 27B 89.1%
  18. 18 MaziyarPanahi calme-2.1-qwen2.5-72b 89.1%
  19. 19 Aya Expanse 32B 89%
  20. 20 tanliboy lambda-qwen2.5-14b-dpo-test 88.8%
  21. 21 Sakalti SakaMoe-3x14B-Instruct 88.8%
  22. 22 zetasepic Qwen2.5-32B-Instruct-abliterated-v2 88.4%
  23. 23 Applied-Innovation-Center Karnak 88.3%
  24. 24 Sakalti Saka-14B 88.3%
  25. 25 QCRI/Fanar-1-9B-Instruct 88.3%
  26. 26 Llama 3.3 70B Instruct 88.3%
  27. 27 maldv Awqward2.5-32B-Instruct 88.3%
  28. 28 maldv Qwentile2.5-32B-Instruct 88.1%
  29. 29 TIGER-Lab Qwen2.5-32B-Instruct-CFT 87.8%
  30. 30 Daemontatox Cogito-R1 87.7%
  31. 31 TarjamaN Pronoia-14b-community 87.6%
  32. 32 Navid-AI Yehia-7B-preview 87.5%
  33. 33 Mushari440 qwen3-8B-SFT 87.3%
  34. 34 huihui-ai Qwen2.5-32B-Instruct-abliterated 87%
  35. 35 v000000 Qwen2.5-Lumen-14B 87%
  36. 36 Gemma 3 12B base model 87%
  37. 37 ALLaM-AI/ALLaM-7B-Instruct-preview 86.9%
  38. 38 v000000 Qwen2.5-14B-Gutenberg-1e-Delta 86.9%
  39. 39 freewheelin free-evo-qwen72b-v0.8-re 86.8%
  40. 40 rombodawg Rombos-LLM-V2.6-Qwen-14b 86.8%
  41. 41 FreedomIntelligence AceGPT-v2-70B-Chat 86.7%
  42. 42 Qwen2.5-32B 86.7%
  43. 43 recoilme recoilme-gemma-2-9B-v0.2 86.5%
  44. 44 tanliboy lambda-qwen2.5-32b-dpo-test 86.5%
  45. 45 Qwen2.5-14B base model 86.5%
  46. 46 Slim205 Barka-9b-it 86.2%
  47. 47 Mushari440 qwen-3-8B-sft-grpo 85.5%
  48. 48 Qwen3-4B base model 84.9%
  49. 49 FreedomIntelligence AceGPT-v2-32B-Chat 84.7%
  50. 50 recoilme recoilme-gemma-2-9B-v0.4 84.5%
  51. 51 FreedomIntelligence AceGPT-v2-70B 84.3%
  52. 52 Ocelotr Qwen3-8B-GAE 84.3%
  53. 53 CohereForAI/aya-23-35B 84.2%
  54. 54 amine-khelif BARS 84%
  55. 55 asas-ai AceGPT-v2-32B 84%
  56. 56 FreedomIntelligence AceGPT-v2-32B 84%
  57. 57 inceptionai/jais-adapted-70b-chat 83.7%
  58. 58 Qwen3-8B base model 83.4%
  59. 59 inceptionai/jais-family-30b-8k-chat 83.1%
  60. 60 silma-ai SILMA-9B-Instruct-v1.0 82.8%
  61. 61 Qwen2-7B-Instruct 82.8%
  62. 62 arcee-ai Arcee-Spark 82.6%
  63. 63 Aya Expanse 8B 82.5%
  64. 64 QCRI/Fanar-1-9B 82.1%
  65. 65 Cran-May T.E-8.1 82.1%
  66. 66 JasperV13 Yehia-7B-DPO-Reasoning-preview 81.9%
  67. 67 airev-ai emirati-14b-v2 81.8%
  68. 68 inceptionai/jais-family-30b-16k-chat 81.6%
  69. 69 huihui-ai Qwen2.5-7B-Instruct-abliterated 81.3%
  70. 70 SeaLLMs SeaLLM-7B-v2.5 80.8%
  71. 71 JasperV13 Yehia-7B-Reasoning-preview 80.8%
  72. 72 Qwen2.5-7B 80.7%
  73. 73 Gemma 2 27B 80.5%
  74. 74 CohereForAI/c4ai-command-r7b-arabic-02-2025 80.5%
  75. 75 Orion-zhen Qwen2.5-7B-Instruct-Uncensored 80.4%
  76. 76 inceptionai/jais-adapted-13b-chat 79.7%
  77. 77 Gemma 3 4B base model 78.3%
  78. 78 Qwen1.5-32B base model 78.1%
  79. 79 SeaLLMs SeaLLMs-v3-7B-Chat 78%
  80. 80 Phi-4 Mini 77.8%
  81. 81 Gemma 2 9B 77.2%
  82. 82 JasperV13 Yehia-7B-SFT-Reasoning-preview 76.6%
  83. 83 FreedomIntelligence AceGPT-v1.5-13B-Chat 76.5%
  84. 84 Sakalti oxyge1-33B 76.5%
  85. 85 Llama 3.1-8B 75.6%
  86. 86 Qwen3-1.7B base model 75.6%
  87. 87 inceptionai/jais-family-13b-chat 75%
  88. 88 inceptionai/jais-adapted-70b 75%
  89. 89 AIDC-AI Marco-LLM-AR-V4 74.4%
  90. 90 huihui-ai Qwen2.5-7B-Instruct-abliterated-v2 74.2%
  91. 91 FreedomIntelligence AceGPT-v2-8B-Chat 74.2%
  92. 92 AIDC-AI Marco-LLM-AR-V2 73.5%
  93. 93 silma-ai SILMA-Kashif-2B-Instruct-v1.0 73.3%
  94. 94 Ocelotr Qwen-8B-SFT 73.3%
  95. 95 Qwen3-30B-A3B-Instruct Jul 2025 72.7%
  96. 96 Qwen2.5-3B 72.5%
  97. 97 Isaak-Carter Josiefied-Qwen2.5-7B-Instruct-abliterated-v2 72.2%
  98. 98 Slim205 Barka-2b-it 71.5%
  99. 99 Qwen2.5-1.5B 70.8%
  100. 100 inceptionai/jais-family-30b-8k 70.7%
  101. 101 Hagrass LLama3-3.2-instruct-trained 69.8%
  102. 102 Qwen2-1.5B-Instruct 68.9%
  103. 103 AIDC-AI Marco-LLM-AR 68.6%
  104. 104 Qwen1.5-7B base model 67.2%
  105. 105 inceptionai/jais-adapted-7b-chat 66%
  106. 106 FreedomIntelligence AceGPT-13B-chat 65.7%
  107. 107 upstage SOLAR-10.7B-v1.0 65.5%
  108. 108 Qwen1.5-14B base model 64.3%
  109. 109 VAGOsolutions SauerkrautLM-Nemo-12b-Instruct 64.3%
  110. 110 FreedomIntelligence AceGPT-7B-chat 63.6%
  111. 111 inceptionai/jais-family-30b-16k 62%
  112. 112 inceptionai/jais-adapted-13b 61.2%
  113. 113 01-ZeroOne SUHAIL-14B-preview 61%
  114. 114 Llama 3.2 3B 58.5%
  115. 115 01-ai/Yi-1.5-9B base model 58.1%
  116. 116 tiiuae/Falcon3-10B-Base base model 58%
  117. 117 Qwen3-14B base model 57.5%
  118. 118 Qwen/Qwen1.5-4B base model 57.2%
  119. 119 01-ai/Yi-1.5-9B-32K base model 56.6%
  120. 120 FreedomIntelligence AceGPT-13B 56.5%
  121. 121 Naseej noon-7b 56.5%
  122. 122 Syed-Hasan-8503 Phi-3-mini-4K-instruct-cpo-simpo 55.8%
  123. 123 speakleash Bielik-11B-v2 55.4%
  124. 124 cognitivecomputations Dolphin3.0-R1-Mistral-24B 54.2%
  125. 125 01-ai/Yi-1.5-34B-32K base model 52.3%
  126. 126 Qwen3-0.6B 51.5%
  127. 127 tiiuae/Falcon3-7B-Base base model 49.9%
  128. 128 FreedomIntelligence AceGPT-7B 49.4%
  129. 129 FreedomIntelligence AceGPT-v1.5-13B 47.7%
  130. 130 Phi-4 47.6%
  131. 131 Sakalti Saka-1.5B 47.5%
  132. 132 Qwen2.5-0.5B 47.3%
  133. 133 Qwen2-0.5B-Instruct 46.9%
  134. 134 INSAIT-Institute BgGPT-7B-Instruct-v0.2 46.7%
  135. 135 qwen3-4b-instruct-2507 40.6%
  136. 136 AIDC-AI Marco-LLM-AR-V3 39.6%
  137. 137 Qwen3-32B 38.9%
  138. 138 Llama 3.2 1B base model 37.1%
  139. 139 inceptionai/jais-family-13b 34.8%
  140. 140 inceptionai/jais-adapted-7b 33.9%
  141. 141 riotu-lab ArabianGPT-01B 32.8%
  142. 142 mobiuslabsgmbh DeepSeek-R1-ReDistill-Llama3-8B-v1.1 32.6%
  143. 143 mobiuslabsgmbh DeepSeek-R1-ReDistill-Qwen-7B-v1.1 32.6%
  144. 144 Qwen/Qwen1.5-1.8B base model 31.5%
  145. 145 Gemma 2 2B IT 31.2%
  146. 146 DeepSeek-R1-Distill-Qwen-14B 31.2%
  147. 147 DeepSeek-R1-Distill-Qwen-32B 31.2%
  148. 148 QwQ-32B 31.2%
  149. 149 stabilityai ar-stablelm-2-base 31.2%
  150. 150 TIGER-Lab Qwen2.5-Math-7B-CFT 30.9%
#EntryScoreSettingDateSource
1 AbdulmalekDS qwen72b-ar-lora 91.4% Sep 2025 Official leaderboard
2 Lina-Z arabic_llm_test 91% Nov 2025 Official leaderboard
3 Mushari440 Qwen3-8B-SFT-V2 90.6% Mar 2026 Official leaderboard
4 Cran-May tempmotacilla-cinerea-0308 90% Mar 2025 Official leaderboard
5 hammh0a Hala-9B 90% Feb 2026 Official leaderboard
6 rombodawg Rombos-LLM-V2.5-Qwen-72b 89.8% Feb 2025 Official leaderboard
7 Applied-Innovation-Center AIC-1 89.8% Sep 2025 Official leaderboard
8 deep-analysis-research D2IL-Arabic-Qwen2.5-72B-Instruct-v0.1 89.7% Jul 2025 Official leaderboard
9 tunny Arabic_Qwen2.5_72B_instruct_finetune_0.1 89.7% Dec 2025 Official leaderboard
10 Lina-Z qwen_arabic_ft 89.7% Jan 2026 Official leaderboard
11 Sakalti Ultiima-72B 89.7% Feb 2025 Official leaderboard
12 deep-analysis-research D2IL-Arabic-Qwen2.5-72B-Instruct-v0.2 89.5% Sep 2025 Official leaderboard
13 Qwen2-72B 89.2% base model Jun 2024 Official leaderboard
14 MaziyarPanahi calme-2.2-qwen2.5-72b 89.2% Jan 2025 Official leaderboard
15 Qwen2.5-72B 89.2% Sep 2024 Official leaderboard
16 AXCXEPT EZO-Qwen2.5-32B-Instruct 89.2% Jan 2025 Official leaderboard
17 Gemma 3 27B 89.1% Mar 2025 Official leaderboard
18 MaziyarPanahi calme-2.1-qwen2.5-72b 89.1% Jan 2025 Official leaderboard
19 Aya Expanse 32B 89% Oct 2024 Official leaderboard
20 tanliboy lambda-qwen2.5-14b-dpo-test 88.8% Jan 2025 Official leaderboard
21 Sakalti SakaMoe-3x14B-Instruct 88.8% Feb 2025 Official leaderboard
22 zetasepic Qwen2.5-32B-Instruct-abliterated-v2 88.4% Mar 2025 Official leaderboard
23 Applied-Innovation-Center Karnak 88.3% Feb 2026 Official leaderboard
24 Sakalti Saka-14B 88.3% Feb 2025 Official leaderboard
25 QCRI/Fanar-1-9B-Instruct 88.3% Jun 2025 Official leaderboard
26 Llama 3.3 70B Instruct 88.3% Dec 2024 Official leaderboard
27 maldv Awqward2.5-32B-Instruct 88.3% Feb 2025 Official leaderboard
28 maldv Qwentile2.5-32B-Instruct 88.1% Feb 2025 Official leaderboard
29 TIGER-Lab Qwen2.5-32B-Instruct-CFT 87.8% Mar 2025 Official leaderboard
30 Daemontatox Cogito-R1 87.7% Feb 2025 Official leaderboard
31 TarjamaN Pronoia-14b-community 87.6% Feb 2025 Official leaderboard
32 Navid-AI Yehia-7B-preview 87.5% Mar 2025 Official leaderboard
33 Mushari440 qwen3-8B-SFT 87.3% Feb 2026 Official leaderboard
34 huihui-ai Qwen2.5-32B-Instruct-abliterated 87% Jan 2025 Official leaderboard
35 v000000 Qwen2.5-Lumen-14B 87% Jan 2025 Official leaderboard
36 Gemma 3 12B 87% base model Mar 2025 Official leaderboard
37 ALLaM-AI/ALLaM-7B-Instruct-preview 86.9% Feb 2025 Official leaderboard
38 v000000 Qwen2.5-14B-Gutenberg-1e-Delta 86.9% Jan 2025 Official leaderboard
39 freewheelin free-evo-qwen72b-v0.8-re 86.8% Jan 2025 Official leaderboard
40 rombodawg Rombos-LLM-V2.6-Qwen-14b 86.8% Jan 2025 Official leaderboard
41 FreedomIntelligence AceGPT-v2-70B-Chat 86.7% Jan 2025 Official leaderboard
42 Qwen2.5-32B 86.7% Sep 2024 Official leaderboard
43 recoilme recoilme-gemma-2-9B-v0.2 86.5% Jan 2025 Official leaderboard
44 tanliboy lambda-qwen2.5-32b-dpo-test 86.5% Jan 2025 Official leaderboard
45 Qwen2.5-14B 86.5% base model Sep 2024 Official leaderboard
46 Slim205 Barka-9b-it 86.2% Jun 2025 Official leaderboard
47 Mushari440 qwen-3-8B-sft-grpo 85.5% Feb 2026 Official leaderboard
48 Qwen3-4B 84.9% base model Apr 2025 Official leaderboard
49 FreedomIntelligence AceGPT-v2-32B-Chat 84.7% Jan 2025 Official leaderboard
50 recoilme recoilme-gemma-2-9B-v0.4 84.5% Jan 2025 Official leaderboard
51 FreedomIntelligence AceGPT-v2-70B 84.3% Jan 2025 Official leaderboard
52 Ocelotr Qwen3-8B-GAE 84.3% Jan 2026 Official leaderboard
53 CohereForAI/aya-23-35B 84.2% Jan 2025 Official leaderboard
54 amine-khelif BARS 84% Sep 2025 Official leaderboard
55 asas-ai AceGPT-v2-32B 84% Jan 2025 Official leaderboard
56 FreedomIntelligence AceGPT-v2-32B 84% Jan 2025 Official leaderboard
57 inceptionai/jais-adapted-70b-chat 83.7% Jan 2025 Official leaderboard
58 Qwen3-8B 83.4% base model Apr 2025 Official leaderboard
59 inceptionai/jais-family-30b-8k-chat 83.1% Jan 2025 Official leaderboard
60 silma-ai SILMA-9B-Instruct-v1.0 82.8% Jan 2025 Official leaderboard
61 Qwen2-7B-Instruct 82.8% Jun 2024 Official leaderboard
62 arcee-ai Arcee-Spark 82.6% Jan 2025 Official leaderboard
63 Aya Expanse 8B 82.5% Oct 2024 Official leaderboard
64 QCRI/Fanar-1-9B 82.1% Jun 2025 Official leaderboard
65 Cran-May T.E-8.1 82.1% Jan 2025 Official leaderboard
66 JasperV13 Yehia-7B-DPO-Reasoning-preview 81.9% May 2025 Official leaderboard
67 airev-ai emirati-14b-v2 81.8% Jan 2025 Official leaderboard
68 inceptionai/jais-family-30b-16k-chat 81.6% Jan 2025 Official leaderboard
69 huihui-ai Qwen2.5-7B-Instruct-abliterated 81.3% Jan 2025 Official leaderboard
70 SeaLLMs SeaLLM-7B-v2.5 80.8% Jan 2025 Official leaderboard
71 JasperV13 Yehia-7B-Reasoning-preview 80.8% May 2025 Official leaderboard
72 Qwen2.5-7B 80.7% Sep 2024 Official leaderboard
73 Gemma 2 27B 80.5% Jun 2024 Official leaderboard
74 CohereForAI/c4ai-command-r7b-arabic-02-2025 80.5% Feb 2025 Official leaderboard
75 Orion-zhen Qwen2.5-7B-Instruct-Uncensored 80.4% Jan 2025 Official leaderboard
76 inceptionai/jais-adapted-13b-chat 79.7% Jan 2025 Official leaderboard
77 Gemma 3 4B 78.3% base model Mar 2025 Official leaderboard
78 Qwen1.5-32B 78.1% base model Feb 2024 Official leaderboard
79 SeaLLMs SeaLLMs-v3-7B-Chat 78% Mar 2025 Official leaderboard
80 Phi-4 Mini 77.8% Feb 2025 Official leaderboard
81 Gemma 2 9B 77.2% Jun 2024 Official leaderboard
82 JasperV13 Yehia-7B-SFT-Reasoning-preview 76.6% May 2025 Official leaderboard
83 FreedomIntelligence AceGPT-v1.5-13B-Chat 76.5% Jan 2025 Official leaderboard
84 Sakalti oxyge1-33B 76.5% Mar 2025 Official leaderboard
85 Llama 3.1-8B 75.6% Jul 2024 Official leaderboard
86 Qwen3-1.7B 75.6% base model Apr 2025 Official leaderboard
87 inceptionai/jais-family-13b-chat 75% Jan 2025 Official leaderboard
88 inceptionai/jais-adapted-70b 75% Jan 2025 Official leaderboard
89 AIDC-AI Marco-LLM-AR-V4 74.4% Mar 2025 Official leaderboard
90 huihui-ai Qwen2.5-7B-Instruct-abliterated-v2 74.2% Jan 2025 Official leaderboard
91 FreedomIntelligence AceGPT-v2-8B-Chat 74.2% Jan 2025 Official leaderboard
92 AIDC-AI Marco-LLM-AR-V2 73.5% Mar 2025 Official leaderboard
93 silma-ai SILMA-Kashif-2B-Instruct-v1.0 73.3% Feb 2025 Official leaderboard
94 Ocelotr Qwen-8B-SFT 73.3% Dec 2025 Official leaderboard
95 Qwen3-30B-A3B-Instruct Jul 2025 72.7% Jul 2025 Official leaderboard
96 Qwen2.5-3B 72.5% Sep 2024 Official leaderboard
97 Isaak-Carter Josiefied-Qwen2.5-7B-Instruct-abliterated-v2 72.2% Jan 2025 Official leaderboard
98 Slim205 Barka-2b-it 71.5% Jun 2025 Official leaderboard
99 Qwen2.5-1.5B 70.8% Sep 2024 Official leaderboard
100 inceptionai/jais-family-30b-8k 70.7% Jan 2025 Official leaderboard
101 Hagrass LLama3-3.2-instruct-trained 69.8% Aug 2025 Official leaderboard
102 Qwen2-1.5B-Instruct 68.9% Jun 2024 Official leaderboard
103 AIDC-AI Marco-LLM-AR 68.6% Feb 2025 Official leaderboard
104 Qwen1.5-7B 67.2% base model Feb 2024 Official leaderboard
105 inceptionai/jais-adapted-7b-chat 66% Jan 2025 Official leaderboard
106 FreedomIntelligence AceGPT-13B-chat 65.7% Jan 2025 Official leaderboard
107 upstage SOLAR-10.7B-v1.0 65.5% Jan 2025 Official leaderboard
108 Qwen1.5-14B 64.3% base model Feb 2024 Official leaderboard
109 VAGOsolutions SauerkrautLM-Nemo-12b-Instruct 64.3% Jan 2025 Official leaderboard
110 FreedomIntelligence AceGPT-7B-chat 63.6% Jan 2025 Official leaderboard
111 inceptionai/jais-family-30b-16k 62% Jan 2025 Official leaderboard
112 inceptionai/jais-adapted-13b 61.2% Jan 2025 Official leaderboard
113 01-ZeroOne SUHAIL-14B-preview 61% Jul 2025 Official leaderboard
114 Llama 3.2 3B 58.5% Sep 2024 Official leaderboard
115 01-ai/Yi-1.5-9B 58.1% base model Jan 2025 Official leaderboard
116 tiiuae/Falcon3-10B-Base 58% base model Feb 2025 Official leaderboard
117 Qwen3-14B 57.5% base model Apr 2025 Official leaderboard
118 Qwen/Qwen1.5-4B 57.2% base model Jan 2025 Official leaderboard
119 01-ai/Yi-1.5-9B-32K 56.6% base model Jan 2025 Official leaderboard
120 FreedomIntelligence AceGPT-13B 56.5% Jan 2025 Official leaderboard
121 Naseej noon-7b 56.5% Mar 2025 Official leaderboard
122 Syed-Hasan-8503 Phi-3-mini-4K-instruct-cpo-simpo 55.8% Jan 2025 Official leaderboard
123 speakleash Bielik-11B-v2 55.4% Jan 2025 Official leaderboard
124 cognitivecomputations Dolphin3.0-R1-Mistral-24B 54.2% Feb 2025 Official leaderboard
125 01-ai/Yi-1.5-34B-32K 52.3% base model Jan 2025 Official leaderboard
126 Qwen3-0.6B 51.5% Apr 2025 Official leaderboard
127 tiiuae/Falcon3-7B-Base 49.9% base model Feb 2025 Official leaderboard
128 FreedomIntelligence AceGPT-7B 49.4% Jan 2025 Official leaderboard
129 FreedomIntelligence AceGPT-v1.5-13B 47.7% Jan 2025 Official leaderboard
130 Phi-4 47.6% Dec 2024 Official leaderboard
131 Sakalti Saka-1.5B 47.5% Feb 2025 Official leaderboard
132 Qwen2.5-0.5B 47.3% Sep 2024 Official leaderboard
133 Qwen2-0.5B-Instruct 46.9% Jun 2024 Official leaderboard
134 INSAIT-Institute BgGPT-7B-Instruct-v0.2 46.7% Jan 2025 Official leaderboard
135 qwen3-4b-instruct-2507 40.6% Aug 2025 Official leaderboard
136 AIDC-AI Marco-LLM-AR-V3 39.6% Mar 2025 Official leaderboard
137 Qwen3-32B 38.9% Apr 2025 Official leaderboard
138 Llama 3.2 1B 37.1% base model Sep 2024 Official leaderboard
139 inceptionai/jais-family-13b 34.8% Jan 2025 Official leaderboard
140 inceptionai/jais-adapted-7b 33.9% Jan 2025 Official leaderboard
141 riotu-lab ArabianGPT-01B 32.8% Oct 2025 Official leaderboard
142 mobiuslabsgmbh DeepSeek-R1-ReDistill-Llama3-8B-v1.1 32.6% Feb 2025 Official leaderboard
143 mobiuslabsgmbh DeepSeek-R1-ReDistill-Qwen-7B-v1.1 32.6% Feb 2025 Official leaderboard
144 Qwen/Qwen1.5-1.8B 31.5% base model Jan 2025 Official leaderboard
145 Gemma 2 2B IT 31.2% Jul 2024 Official leaderboard
146 DeepSeek-R1-Distill-Qwen-14B 31.2% Jan 2025 Official leaderboard
147 DeepSeek-R1-Distill-Qwen-32B 31.2% Jan 2025 Official leaderboard
148 QwQ-32B 31.2% Mar 2025 Official leaderboard
149 stabilityai ar-stablelm-2-base 31.2% Jun 2025 Official leaderboard
150 TIGER-Lab Qwen2.5-Math-7B-CFT 30.9% Mar 2025 Official leaderboard

Labs report scores for 1 model in their cards. We plot every source in the charts below.

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.

Lowest effort to highest, light to dark
Qwen2.5-72B
base model
89.2%
Gemma 3 27B
base model
89.1%
Qwen2.5-32B
base model
86.7%
Qwen2.5-14B
base model
86.5%
Qwen3-4B
base model
84.9%
Qwen3-8B
base model
83.4%
Qwen2.5-7B
base model
80.7%
Gemma 2 27B
base model
80.5%
Llama 3.1-8B
base model
75.6%
Qwen2.5-3B
base model
72.5%
Qwen2.5-1.5B
base model
70.8%
Qwen3-14B
base model
57.5%
ModelEffortHarnessScoreCost per taskOutput tokens
Qwen2-72B base model 89.2%
Qwen2.5-72B 89.2%
Gemma 3 27B 89.1%
Aya Expanse 32B 89%
Qwen2.5-72B base model 88.5%
Llama 3.3 70B Instruct 88.3%
Gemma 3 27B base model 87.8%
Gemma 3 12B base model 87%
Qwen2.5-32B 86.7%
Qwen2.5-14B base model 86.5%
Qwen2.5-14B 86.1%
Qwen3-4B base model 84.9%
Qwen3-8B base model 83.4%
Qwen2.5-32B base model 83%
Qwen2-7B-Instruct 82.8%
Aya Expanse 8B 82.5%
Qwen2.5-7B 80.7%
Gemma 2 27B 80.5%
Gemma 3 4B base model 78.3%
Qwen1.5-32B base model 78.1%
Phi-4 Mini 77.8%
Qwen2.5-7B base model 77.6%
Gemma 2 9B 77.2%
Llama 3.1-8B 75.6%
Qwen3-1.7B base model 75.6%
Qwen3-30B-A3B-Instruct Jul 2025 72.7%
Qwen2.5-3B 72.5%
Llama 3.1-8B base model 72%
Qwen2.5-1.5B 70.8%
Qwen2-1.5B-Instruct 68.9%
Qwen1.5-7B base model 67.2%
Qwen1.5-14B base model 64.3%
Llama 3.2 3B 58.5%
Qwen3-14B base model 57.5%
Qwen3-8B 55.3%
Qwen3-4B 52.8%
Qwen3-0.6B 51.5%
Qwen3-14B 51.1%
Phi-4 47.6%
Qwen2.5-0.5B 47.3%
Qwen2-0.5B-Instruct 46.9%
Qwen2.5-3B base model 43.8%
qwen3-4b-instruct-2507 40.6%
Qwen3-32B 38.9%
Llama 3.2 1B base model 37.1%
Gemma 2 27B base model 31.2%
Gemma 2 2B IT 31.2%
Qwen2.5-0.5B base model 31.2%
DeepSeek-R1-Distill-Qwen-14B 31.2%
DeepSeek-R1-Distill-Qwen-32B 31.2%
QwQ-32B 31.2%
Qwen2.5-1.5B base model 30.2%

2 models have held number one since its release in Mar 2024

Lab claims count from the day the card came out, independent runs from the model’s release.

Qwen
Own leaderboard Passed 85% in Jun 2024, by Qwen2-72B
Qwen2-72B 89.2%

The best score passed 85% in Jun 2024

All the scores we have, one dot each, so a model tested at several settings appears more than once.

On the benchmark’s own leaderboard Reported by the lab in its model card Best so far
0% 20% 40% 60% 80% 100% 2024 2025 2026 Near the ceiling (85%) Qwen2-72B Qwen1.5-32B 0% 20% 40% 60% 80% 100% 2024 2025 2026 Near the ceiling (85%) Qwen2-72B Qwen1.5-32B

Labs dropped it from launch posts once the best score passed 85%

Launch posts from every lab that has listed it, one tick per post. Tall colored ticks include it, and short gray ones leave it out.

Technology Innovation Institute
DateLaunch postScoreAs the post writes it
21 May 2025 Falcon-Arabic-7B 82.5 AraTrust

One system card reports it

Besides launch posts, we read the system cards, model cards and technical reports labs publish with a release. They report far more, including safety and frontier-risk tests.

DateDocumentScoreAs the document writes itSection
8 Nov 2025 Falcon-Arabic-7B technical report 82.6 AraTrust (Table 6) Safety

More in Safety and alignment

All 444 in Safety and alignment
  • FORTRESS In use Paired adversarial and benign national-security misuse prompts. Rubric-graded risk score on adversarial and over-refusal score on benign prompts. Claude Opus 5.5 5.78%
  • StrongREJECT In use Forbidden-prompt jailbreak benchmark with a rubric grader that scores how useful a non-refusal actually is. Inkling 98.6%
  • AgentDojo In use Agents do tool-use tasks while tools and data carry prompt injections. Attack success rate counts hijacked runs. Grok 4 Fast 0%
  • MASK In use Measures whether a model contradicts its own elicited beliefs under pressure to lie. Honesty separated from accuracy. Muse Spark Open 98.1%
  • Gray Swan IPI Arena benchmark In use Indirect prompt-injection attacks from public red-teaming competitions replayed on tool, coding and browser agents. Attack success rate at k attempts. GPT-6 Astra 8.50%
  • Agentic misalignment suite (Anthropic) In use Agents in simulated company settings face threats or goal conflicts that invite blackmail, leaking or sabotage. Score is the misaligned action rate. Claude Sonnet 4.5 0%