跳到主要内容
ORIEL
媒体
710

媒体竞技场

11

图像、视频、语音与音乐模型的人类偏好排名。Elo 带 95% 置信区间,区间重叠即代表名次差异不显著。

文生图

Elo · 162
  1. 1GPT Image 2.5 Sunburst (max)1,196±9
  2. 2GPT Image 2.5 Flare (max)1,190±9
  3. 3GPT Image 2 (high)1,171±9
  4. 4Grok Imagine Image 2.01,157±11
  5. 5MAI-Image-2.61,148±10
  6. 6Nano Banana 2 (Gemini 3.1 Flash Image)1,123±8
  7. 7Muse Image1,111±10
  8. 8GPT Image 1.5 (high)1,104±9
  9. 9MAI-Image-2.51,102±8
  10. 10MAI-Image-2.6-Flash1,102±9
  11. 11Nano Banana Pro (Gemini 3 Pro Image)1,101±8
  12. 12MAI-Image-2.5-Pro1,100±8
  13. 13Nano Banana 2 Lite (Gemini 3.1 Flash Lite Image)1,093±8
  14. 14Qwen-Image-3.0-Pro1,088±10
  15. 15Seedream 5.0 Pro1,078±8
  16. 16Qwen-Image-3.01,074±10
  17. 17grok-imagine-image-quality1,045±7
  18. 18Qwen-Image-2.11,035±10
  19. 19MAI-Image-2.5-Flash1,033±9
  20. 20Qwen Image 2.0 Pro (2026-04-22)1,029±9
  21. 21Seedream 4.01,028±10
  22. 22Wan 2.6 Image1,027±9
  23. 23FLUX.2 [flex]1,026±9
  24. 24Krea 2 Large1,025±7
  25. 25Seedream 4.51,021±9
  26. 26HiDream-O1-Image-1.51,021±8
  27. 27Recraft V4.1 Utility1,021±9
  28. 28FLUX.2 [max]1,021±8
  29. 29grok-imagine-image1,020±9
  30. 30Krea 2 Medium Turbo1,018±9
  31. 31Recraft V4.1 Utility Pro1,017±7
  32. 32Wan2.6 Text to Image1,014±9
  33. 33Luma UNI 1 Max1,014±9
  34. 34Krea 2 Medium1,012±7
  35. 35GPT Image 1 (high)1,011±9
  36. 36Ideogram 4.0 (Quality)1,010±7
  37. 37Seedream 5.0 Lite1,009±9
  38. 38MAI-Image-21,008±9
  39. 39Ideogram 4.01,003±8
  40. 40FLUX.2 [pro]1,002±8
  41. 41FLUX.2 [dev]1,000±0
  42. 42FLUX.2 [dev] Turbo999±9
  43. 43Qwen Image Max 2512998±10
  44. 44Imagen 4 Ultra997±10
  45. 45P-Image-Ideogram (High)997±10
  46. 46Ming-Image-0.1-Design996±8
  47. 47Cosmos3-Super-Text2Image (agentic)995±8
  48. 48Luma UNI 1991±9
  49. 49Recraft V4.1990±9
  50. 50MAI-Image-2-Efficient989±10
  51. 51Wan 2.7989±9
  52. 52Recraft V4 Pro986±9
  53. 53Ideogram 4.0 Fast (Quality)985±9
  54. 54Nano Banana (Gemini 2.5 Flash Image)985±9
  55. 55Cosmos3-Super-Text2Image985±8
  56. 56P-Image-Ideogram (Very High)985±11
  57. 57Recraft V4.1 Pro984±9
  58. 58FLUX.2 [dev] Flash984±9
  59. 59Wan 2.7 Pro983±9
  60. 60Recraft V4982±9
  61. 61Ideogram 4.0 Instant982±9
  62. 62HiDream-O1-Image979±8
  63. 63P-Image-Ideogram (Medium)974±10
  64. 64Cosmos3-Super-Text2Image-4Step970±8
  65. 65ImagineArt 2.0970±9
  66. 66Ideogram 4.0 Fast969±9
  67. 67P-Image-Ideogram (Low)965±10
  68. 68HunyuanImage 3.0 Instruct (Fal)963±9
  69. 69Wan 2.5 Preview961±9
  70. 70Qwen Image 2.0 (2026-03-03)960±8
  71. 71Seedream 3.0958±9
  72. 72ImagineArt 1.5 Preview957±7
  73. 73HunyuanImage 3.0 (Fal)945±9
  74. 74Vivago 2.1944±9
  75. 75Kolors 2.1943±9
  76. 76Z-Image Turbo940±9
  77. 77image-1940±9
  78. 78FLUX.2 [klein] 9B940±8
  79. 79Qwen Image Plus 2601935±7
  80. 80FLUX.1 Kontext [max]933±9
  81. 81Eigen Image923±9
  82. 82ERNIE Image Turbo923±8
  83. 83Vivago 2.0923±9
  84. 84Vidu Q2919±7
  85. 85P-Image-Ideogram (Very Low)916±10
  86. 86Imagen 3 (v002)915±9
  87. 87GPT Image 1 Mini (medium)914±7
  88. 88Kling Image 3.0 Omni914±7
  89. 89Dreamina 3.1913±7
  90. 90Lucid Origin Ultra913±7
  91. 91ERNIE Image912±8
  92. 92Reve Image (Halfmoon)912±9
  93. 93Imagen 4 Standard909±10
  94. 94FLUX.1 Kontext [pro]908±7
  95. 95FLUX.2 [klein] Base 9B902±9
  96. 96Lucid Origin Fast899±7
  97. 97Ideogram 3.0897±8
  98. 98FLUX1.1 [pro] Ultra893±7
  99. 99FLUX1.1 [pro]890±7
  100. 100FLUX.1 [pro]890±8
  101. 101GLM-Image889±7
  102. 102Qwen Image887±7
  103. 103HunyuanImage 2.1886±9
  104. 104Luma Photon885±7
  105. 105P-Image883±8
  106. 106Imagen 4 Fast882±10
  107. 107Ideogram v2882±8
  108. 108FIBO879±8
  109. 109HiDream-O1-Image-Dev878±8
  110. 110Midjourney v7 Alpha878±10
  111. 111Z-Image Base875±8
  112. 112Midjourney v6874±8
  113. 113Infinity 8B873±9
  114. 114HiDream-I1-Dev873±10
  115. 115Ideogram v2 Turbo870±8
  116. 116HiDream-I1-Fast868±9
  117. 117Recraft V3868±7
  118. 118SRPO864±8
  119. 119FLUX.2 [klein] 4B864±9
  120. 120MAI Image 1864±9
  121. 121Image-01860±9
  122. 122LongCat Image860±8
  123. 123Midjourney v6.1847±8
  124. 124FLUX.1 Krea [dev]844±9
  125. 125Phoenix 1.0 Ultra842±9
  126. 126FLUX.1 [dev]841±7
  127. 127Stable Diffusion 3.5 Large839±9
  128. 128Stable Diffusion 3 Large837±8
  129. 129Phoenix 0.9 Ultra832±7
  130. 130Ideogram v2a Turbo831±9
  131. 131Stable Diffusion 3.5 Large Turbo828±9
  132. 132FIBO Lite827±11
  133. 133Phoenix 1.0 Fast823±8
  134. 134Ideogram v2a819±10
  135. 135Krea 1819±9
  136. 136Playground v3 (beta)809±8
  137. 137FLUX.1 [schnell]802±7
  138. 138Step Image Edit 2797±11
  139. 139Luma Photon Flash797±7
  140. 140FLUX.2 [klein] Base 4B793±10
  141. 141Runway Gen-4 Image791±9
  142. 142Lumina Image v2781±9
  143. 143DALLE 3776±8
  144. 144Playground v2.5776±8
  145. 145DALLE 3 HD774±8
  146. 146Recraft 20B772±7
  147. 147Stable Diffusion 3.5 Medium758±10
  148. 148Sana Sprint 1.6B753±10
  149. 149Amazon Titan G1 v2 (Standard)730±9
  150. 150OmniGen V2726±9
  151. 151Stable Diffusion 3 Large Turbo720±9
  152. 152Stable Diffusion 3 Medium719±9
  153. 153SDXL Lightning719±9
  154. 154Stable Diffusion 1.6718±9
  155. 155Amazon Titan G1 (Standard)707±9
  156. 156Bagel704±9
  157. 157Bria 3.2704±10
  158. 158Stable Diffusion XL 1.0677±9
  159. 159Stable Diffusion 2.1548±11
  160. 160DALLE 2542±11
  161. 161Janus Pro532±12
  162. 162Stable Diffusion 1.5455±13

Elo 455–1196

误差条重叠的模型,排名先后没有统计意义

图像编辑

Elo · 77
  1. 1GPT Image 2.5 Sunburst (max)1,180±8
  2. 2GPT Image 2.5 Flare (max)1,161±8
  3. 3MAI-Image-2.61,134±8
  4. 4MAI-Image-2.6-Flash1,124±8
  5. 5GPT Image 2 (high)1,121±8
  6. 6Muse Image1,117±8
  7. 7MAI-Image-2.51,113±9
  8. 8Nano Banana 2 (Gemini 3.1 Flash Image)1,107±8
  9. 9Seedream 5.0 Pro1,106±9
  10. 10MAI-Image-2.5-Pro1,105±8
  11. 11Grok Imagine Image 2.01,105±9
  12. 12GPT Image 1.5 (high)1,102±8
  13. 13Nano Banana Pro (Gemini 3 Pro Image)1,098±8
  14. 14MAI-Image-2.5-Flash1,088±8
  15. 15grok-imagine-image-quality1,082±9
  16. 16Qwen-Image-3.0-Pro1,077±8
  17. 17Luma UNI 1 Max1,068±9
  18. 18grok-imagine-image1,067±9
  19. 19Qwen-Image-2.11,067±10
  20. 20HiDream-O1-Edit-1.51,065±9
  21. 21Seedream 5.0 Lite1,057±9
  22. 22Qwen-Image-3.01,052±9
  23. 23Nano Banana 2 Lite (Gemini 3.1 Flash Lite Image)1,042±9
  24. 24DreamImage 2.01,041±9
  25. 25Wan 2.7 Pro1,041±9
  26. 26Luma UNI 11,039±8
  27. 27Wan 2.6 Image1,038±9
  28. 28Wan 2.71,036±9
  29. 29Seedream 4.51,034±9
  30. 30HunyuanImage 3.0 Instruct (Fal)1,028±9
  31. 31Qwen Image Edit Plus 25111,023±8
  32. 32Qwen Image 2.0 Pro (2026-04-22)1,016±8
  33. 33FLUX.2 [klein] 9B1,014±9
  34. 34Seedream 4.01,013±8
  35. 35Eigen Image1,013±10
  36. 36Agnes Image 2.0 Flash1,011±11
  37. 37Reve V1 (December)1,009±8
  38. 38FLUX.2 [pro]1,007±7
  39. 39FLUX.2 [flex]1,007±7
  40. 40Step Image Edit 21,005±9
  41. 41Kling Image 3.01,001±8
  42. 42FLUX.2 [dev] Turbo1,000±8
  43. 43FLUX.2 [dev]1,000±0
  44. 44Qwen Image Edit Max 26011,000±8
  45. 45FLUX.2 [dev] Flash996±8
  46. 46FLUX.2 [max]995±9
  47. 47Nano Banana (Gemini 2.5 Flash Image)985±9
  48. 48Wan 2.5 Preview983±10
  49. 49GPT Image 1 (high)982±7
  50. 50P-Image-Edit981±9
  51. 51Qwen Image Edit Plus 2509981±7
  52. 52FLUX.2 [klein] Base 9B973±10
  53. 53ImagineArt 2.0968±12
  54. 54Vidu Q2966±7
  55. 55Qwen Image 2.0 (2026-03-03)962±9
  56. 56Kling Image 3.0 Omni953±9
  57. 57HiDream-O1-Image952±9
  58. 58FLUX.2 [klein] 4B949±7
  59. 59Kling Image O1946±9
  60. 60LongCat Image938±8
  61. 61FIBO Edit930±9
  62. 62FLUX.1 Kontext [max]927±7
  63. 63Step1X-Edit-v1p2922±10
  64. 64GPT Image 1 Mini (medium)922±7
  65. 65Qwen Image Edit918±7
  66. 66SeedEdit 3.0917±7
  67. 67FLUX.1 Kontext [pro]912±7
  68. 68HiDream-O1-Image-Dev887±10
  69. 69FLUX.2 [klein] Base 4B868±10
  70. 70FLUX.1 Kontext [dev]857±8
  71. 71HiDream-E1.1808±11
  72. 72step1x-edit-v1p2-preview803±9
  73. 73GLM-Image781±9
  74. 74Bagel771±8
  75. 75OmniGen V2769±8
  76. 76Step1X-Edit674±10
  77. 77HiDream-E1-Full669±18

Elo 669–1180

误差条重叠的模型,排名先后没有统计意义

文生视频

Elo · 81
  1. 1Wan 3.01,335±12
  2. 2Gemini Omni Flash1,332±10
  3. 3MiniMax H31,302±10
  4. 4HappyHorse-1.01,286±8
  5. 5HappyHorse-1.11,272±9
  6. 6Dreamina Seedance 2.0 720p1,259±8
  7. 7Wan2.7-2606121,243±9
  8. 8grok-imagine-video1,235±8
  9. 9PixVerse V5.61,230±8
  10. 10PixVerse V61,230±8
  11. 11Kling 3.0 Omni 1080p (Pro)1,230±8
  12. 12Kling 3.0 1080p (Pro)1,230±8
  13. 13Vidu Q3 Pro1,225±8
  14. 14Bach-1.0 Preview1,222±9
  15. 15PixVerse V5.51,219±10
  16. 16LTX-2.5 Fast1,218±11
  17. 17Veo 3.1 Lite1,215±10
  18. 18Kling 3.0 Omni 720p (Standard)1,215±8
  19. 19Runway Gen-4.51,214±8
  20. 20Veo 3.11,214±8
  21. 21Veo 31,211±8
  22. 22SkyReels V41,211±9
  23. 23Wan 2.71,210±9
  24. 24Kling 3.0 720p (Standard)1,210±8
  25. 25LTX-2.5 Pro1,209±11
  26. 26Veo 3.1 Fast1,208±8
  27. 27Kling 2.5 Turbo 1080p1,203±8
  28. 28Wan 2.61,197±9
  29. 29Kling 2.6 Pro (January)1,196±8
  30. 30Kling O1 Pro (January)1,193±8
  31. 31Kling O1 Standard (January)1,185±8
  32. 32Ray 31,185±8
  33. 33Kling 2.6 Standard (January)1,184±8
  34. 34Sora 2 Pro1,179±9
  35. 35Seedance 1.5 pro1,176±8
  36. 36PixVerse V51,176±8
  37. 37Hailuo 2.31,176±8
  38. 38Veo 3 Fast Preview1,173±9
  39. 39Vidu Q21,171±8
  40. 40Hailuo 02 Standard1,170±9
  41. 41Sora 2 (December)1,168±9
  42. 42Wan 2.5 Preview1,160±8
  43. 43Hailuo 02 Pro1,154±9
  44. 44Seedance 1.01,136±8
  45. 45LTX-2 Pro1,127±8
  46. 46LTX-2 Fast1,125±8
  47. 47LTX-2.3 Pro1,124±9
  48. 48LTX-2.3 Fast1,121±9
  49. 49Wan 2.2 A14B1,116±8
  50. 50Veo 21,108±8
  51. 51Vidu Q3 Turbo1,105±11
  52. 52Kling 2.01,092±9
  53. 53Pika 2.51,086±8
  54. 54Seedance 1.0 Mini1,084±9
  55. 55PixVerse V4.51,083±10
  56. 56P-Video1,059±9
  57. 57Agnes-Video-V2.01,054±10
  58. 58Marey1,047±9
  59. 59Kling 1.5 Pro1,032±9
  60. 60T2V-011,026±8
  61. 61T2V-01-Director1,024±9
  62. 62HunyuanVideo-1.5 (Fal)1,020±9
  63. 63Pika 2.01,020±9
  64. 64Kling 1.6 Pro1,018±9
  65. 65Wan 2.1 14B1,018±9
  66. 66Kling 1.6 Standard1,010±9
  67. 67Vidu Q11,002±11
  68. 68Mochi 11,000±0
  69. 69Hunyuan Video (Fal)992±9
  70. 70Runway Gen 3 Alpha986±8
  71. 71Krea Realtime982±10
  72. 72Ray 1979±9
  73. 73Wan 2.2 5B950±10
  74. 74Pika 2.2949±10
  75. 75Kling 1.0949±9
  76. 76Haiper 2.0945±9
  77. 77Ray 2942±9
  78. 78Pika 1.5931±9
  79. 79Step-Video-T2V904±11
  80. 80CogVideoX-5B805±10
  81. 81Pyramid Flow778±11

Elo 778–1335

误差条重叠的模型,排名先后没有统计意义

图生视频

Elo · 76
  1. 1Gemini Omni Flash1,369±11
  2. 2Wan 3.01,362±13
  3. 3Bach 1.0 Pro1,362±12
  4. 4MiniMax H31,357±11
  5. 5PixVerse V61,340±10
  6. 6Dreamina Seedance 2.0 720p1,339±10
  7. 7grok-imagine-video-1.51,331±11
  8. 8grok-imagine-video1,329±9
  9. 9HappyHorse-1.11,315±10
  10. 10Kling 2.5 Turbo 1080p1,299±9
  11. 11HappyHorse-1.01,295±10
  12. 12Vidu Q3 Pro1,291±9
  13. 13PixVerse V5.61,287±10
  14. 14SkyReels V41,283±11
  15. 15Kling 3.0 Omni 1080p (Pro)1,280±10
  16. 16Kling 3.0 1080p (Pro)1,278±9
  17. 17PixVerse V51,277±9
  18. 18PixVerse V5.51,275±10
  19. 19Cosmos3-Super-Image2Video-4Step1,274±12
  20. 20Kling 3.0 Omni 720p (Standard)1,271±10
  21. 21Kling 3.0 720p (Standard)1,270±9
  22. 22Wan 2.71,270±11
  23. 23Kling 2.6 Pro (January)1,266±10
  24. 24Kling 2.6 Standard (January)1,265±10
  25. 25Runway Gen-4.51,264±10
  26. 26Veo 3.1 Fast1,259±10
  27. 27Hailuo 2.3 Fast1,257±10
  28. 28Seedance 1.5 pro1,256±9
  29. 29Hailuo 2.31,255±9
  30. 30Cosmos3-Super-Image2Video1,255±11
  31. 31Veo 3.11,253±10
  32. 32Hailuo 02 Pro1,250±11
  33. 33Hailuo 02 Standard1,247±11
  34. 34Wan 2.5 Preview1,246±9
  35. 35Veo 3.1 Lite1,241±13
  36. 36Vidu Q2 Turbo1,237±10
  37. 37Seedance 1.01,235±10
  38. 38Vidu Q2 Pro1,234±10
  39. 39Vidu Q3 Turbo1,230±12
  40. 40Veo 31,225±10
  41. 41Wan 2.61,222±11
  42. 42Ray 31,216±10
  43. 43LTX-2.5 Fast1,214±12
  44. 44Kling 2.1 Master1,208±12
  45. 45Hailuo 02 Fast1,208±11
  46. 46Pika 2.51,203±10
  47. 47LTX-2 Pro1,201±9
  48. 48LTX-2.5 Pro1,193±12
  49. 49LTX-2 Fast1,191±10
  50. 50Kling 2.1 Pro1,185±11
  51. 51Kling 2.01,180±10
  52. 52Vivago 2.01,177±10
  53. 53Kling 2.1 Standard1,174±11
  54. 54Midjourney V11,166±11
  55. 55LTX-2.3 Pro1,159±10
  56. 56PixVerse V4.51,154±12
  57. 57LTX-2.3 Fast1,150±11
  58. 58Seedance 1.0 Mini1,139±11
  59. 59Kling 1.6 Pro1,137±10
  60. 60HunyuanVideo-1.5 (Fal)1,134±11
  61. 61P-Video1,130±11
  62. 62Agnes-Video-V2.01,127±11
  63. 63Wan 2.2 A14B1,107±10
  64. 64Veo 21,100±11
  65. 65Runway Gen 41,088±11
  66. 66I2V-01-Director1,049±11
  67. 67Vidu Q11,032±13
  68. 68LTX Video v0.9.7 13B1,030±13
  69. 69Motion 2.01,027±13
  70. 70Runway Gen 3 Alpha1,016±12
  71. 71Pika 2.21,014±12
  72. 72Marey1,007±12
  73. 73Runway Gen 3 Alpha Turbo1,004±12
  74. 74Wan 2.1 14B1,000±0
  75. 75Wan 2.2 5B993±12
  76. 76Hunyuan Video (Fal)921±14

Elo 921–1369

误差条重叠的模型,排名先后没有统计意义

文生视频(含声)

Elo · 36
  1. 1Gemini Omni Flash1,233±7
  2. 2Wan 3.01,229±9
  3. 3Minimax H3 Max (post-trained by fal)1,227±9
  4. 4MiniMax H31,220±8
  5. 5Dreamina Seedance 2.0 720p1,210±6
  6. 6MAGI-2 Preview (0912)1,156±9
  7. 7Wan2.7-2606121,149±6
  8. 8HappyHorse-1.11,147±6
  9. 9HappyHorse-1.01,119±7
  10. 10Wan 2.71,098±7
  11. 11SkyReels V41,095±7
  12. 12Kling 3.0 1080p (Pro)1,095±6
  13. 13Kling 3.0 720p (Standard)1,089±6
  14. 14Veo 3.11,088±7
  15. 15Veo 3.1 Fast1,085±6
  16. 16Veo 3.1 Lite1,083±7
  17. 17Sora 2 (December)1,083±10
  18. 18Kling 3.0 Omni 720p (Standard)1,082±7
  19. 19Vidu Q3 Pro1,078±6
  20. 20Agnes-Video-2.51,076±9
  21. 21PixVerse V61,076±7
  22. 22Kling 3.0 Omni 1080p (Pro)1,075±7
  23. 23grok-imagine-video1,065±6
  24. 24Sora 2 Pro1,059±10
  25. 25LTX-2.5 Fast1,055±9
  26. 26LTX-2.5 Pro1,053±9
  27. 27Wan 2.61,030±7
  28. 28Vidu Q3 Turbo1,025±9
  29. 29Seedance 1.5 pro1,000±0
  30. 30LTX-2.3 Fast975±7
  31. 31Kling 2.6 Pro (January)974±8
  32. 32LTX-2.3 Pro958±7
  33. 33PixVerse V5.6952±8
  34. 34LTX-2 Fast924±9
  35. 35LTX-2 Pro905±9
  36. 36Agnes-Video-V2.0897±9

Elo 897–1233

误差条重叠的模型,排名先后没有统计意义

图生视频(含声)

Elo · 35
  1. 1Minimax H3 Max (post-trained by fal)1,194±9
  2. 2MiniMax H31,181±8
  3. 3Gemini Omni Flash1,178±7
  4. 4Dreamina Seedance 2.0 720p1,175±7
  5. 5HiDream-O1-Video1,174±10
  6. 6Wan 3.01,164±9
  7. 7HappyHorse-1.11,104±7
  8. 8grok-imagine-video-1.51,099±8
  9. 9MAGI-2 Preview1,096±7
  10. 10HappyHorse-1.01,082±7
  11. 11Veo 3.11,081±7
  12. 12Wan 2.71,077±8
  13. 13grok-imagine-video1,072±7
  14. 14Veo 3.1 Lite1,072±8
  15. 15PixVerse V61,071±7
  16. 16SkyReels V41,069±8
  17. 17Veo 3.1 Fast1,066±6
  18. 18Vidu Q3 Pro1,056±6
  19. 19Kling 3.0 1080p (Pro)1,054±6
  20. 20Kling 3.0 720p (Standard)1,051±7
  21. 21Kling 3.0 Omni 1080p (Pro)1,045±7
  22. 22LTX-2.5 Fast1,036±9
  23. 23Kling 3.0 Omni 720p (Standard)1,035±7
  24. 24Agnes-Video-2.51,031±9
  25. 25Vidu Q3 Turbo1,029±9
  26. 26LTX-2.5 Pro1,004±10
  27. 27Seedance 1.5 pro1,000±0
  28. 28Kling 2.6 Pro (January)988±8
  29. 29PixVerse V5.6963±8
  30. 30LTX-2.3 Pro948±8
  31. 31LTX-2.3 Fast947±8
  32. 32Agnes-Video-V2.0924±9
  33. 33LTX-2 Fast923±9
  34. 34Wan 2.6889±9
  35. 35LTX-2 Pro863±10

Elo 863–1194

误差条重叠的模型,排名先后没有统计意义

文本转语音

Elo · 91
  1. 1Sonic 3.61,279±17
  2. 2Gemini 3.8 Flash TTS1,265±17
  3. 3Qwen-Audio-3.0-TTS-Plus1,259±16
  4. 4Realtime TTS-21,246±18
  5. 5Simba 3.21,239±14
  6. 6Gemini 3.8 Flash-Lite TTS1,239±16
  7. 7Luna TTS1,229±14
  8. 8Realtime TTS-2 Flash1,210±15
  9. 9Breeze TTS 21,205±16
  10. 10Gemini 3.1 Flash TTS1,202±12
  11. 11StepAudio 2.5 TTS (Aug 2026)1,199±16
  12. 12v3 Conversational1,196±15
  13. 13Sonic 3.51,185±12
  14. 14Lightning V3.1 Pro (Jul 2026)1,174±14
  15. 15Soniox TTS Real-Time v21,173±12
  16. 16Speech 2.8 HD1,170±11
  17. 17Eleven v31,169±11
  18. 18Falcon 21,159±15
  19. 19Speech 2.8 Turbo1,149±11
  20. 20Gradium TTS (Aug 2026)1,149±14
  21. 21Fish Audio S2.1 Pro1,140±13
  22. 22Step TTS 2 (Mar 2026)1,130±13
  23. 23Async Flash v1.51,130±13
  24. 24SpaceXAI TTS1,130±15
  25. 25Speech 2.6 HD1,129±12
  26. 26Azure HD 2.51,128±12
  27. 27Fish Audio S2 Pro1,122±13
  28. 28Simba 3.01,119±12
  29. 29Speech 2.6 Turbo1,115±11
  30. 30Async Pro v1.01,115±13
  31. 31Speech-02-HD1,111±11
  32. 32TTS-1 HD1,101±12
  33. 33Chatterbox HD1,098±13
  34. 34Speech 2.61,096±13
  35. 35Turbo v2.51,095±11
  36. 36Step Audio EditX (Mar 2026)1,095±13
  37. 37Multilingual v21,093±10
  38. 38TTS-11,087±11
  39. 39Speech-02-Turbo1,085±12
  40. 40Gemini 2.5 Flash Lite TTS1,081±12
  41. 41Voxtral TTS1,078±13
  42. 42OpenAudio S11,077±11
  43. 43Studio1,075±12
  44. 44Flash v2.51,074±11
  45. 45GPT-Realtime-21,074±13
  46. 46Journey1,073±13
  47. 47Maya 2 Flash1,067±14
  48. 48Kokoro 82M v1.01,063±11
  49. 49Magpie-Multilingual 357M (Feb 2026)1,063±13
  50. 50T2A-01-HD1,061±11
  51. 51Polly Generative1,061±12
  52. 52Sonic 31,057±12
  53. 53Coda1,055±13
  54. 54Chirp 3: HD1,053±12
  55. 55Gemini 2.5 Flash TTS (Dec 2025)1,052±12
  56. 56Async Flash v1.01,050±12
  57. 57Octave 21,050±12
  58. 58MiMo-V2.5-TTS1,049±13
  59. 59Maya 2 Global1,047±14
  60. 60Maya11,044±12
  61. 61OpenAudio S1 Mini1,041±20
  62. 62Bland Speech v31,034±13
  63. 63MAI-Voice-11,032±13
  64. 64Higgs Audio V3 TTS1,032±13
  65. 65Polly Long-Form1,032±13
  66. 66Azure Neural1,032±19
  67. 67Octave TTS1,028±12
  68. 68T2A-01-Turbo1,027±11
  69. 69Gemini 2.5 Pro (Dec 2025)1,026±13
  70. 70Chatterbox1,023±11
  71. 71MiMo-V2-TTS1,022±14
  72. 72Lightning v3.11,020±13
  73. 73Raon SpeechLM1,002±14
  74. 74Zonos-v0.11,000±0
  75. 75Murf Speech Gen 2983±12
  76. 76VibeVoice 1.5B950±14
  77. 77OpenVoice v2949±13
  78. 78Qwen3 TTS Flash943±14
  79. 79Neuphonic TTS937±14
  80. 80Kugel 3933±13
  81. 81Qwen3 TTS927±13
  82. 82WaveNet917±12
  83. 83XTTS v2916±15
  84. 84Mist V2908±15
  85. 85Neural2895±12
  86. 86StyleTTS 2892±15
  87. 87Polly Neural890±14
  88. 88Noiz TTS883±15
  89. 89Standard882±12
  90. 90MetaVoice v1845±18
  91. 91Polly Standard829±16

Elo 829–1279

误差条重叠的模型,排名先后没有统计意义

语音对话

得分 · 44
  1. 1StepAudio 3 Realtime1
  2. 2Qwen3.5 Omni Plus Realtime0.99
  3. 3Qwen Audio 3.0 Realtime Plus0.99
  4. 4Step-Audio R1.1 (Realtime)0.98
  5. 5Gemini 3.1 Flash Audio Preview (Mar 2026) - High0.97
  6. 6Grok Voice Think Fast 1.00.97
  7. 7GPT-Realtime-2 High0.97
  8. 8Grok Voice Think Fast 2.0 High0.97
  9. 9Qwen Audio 3.0 Realtime Flash0.96
  10. 10GPT-Realtime-2.1 High0.96
  11. 11Grok Voice Agent0.93
  12. 12GPT-Realtime-2 Medium0.93
  13. 13Gemini 3.8 Live0.92
  14. 14Gemini 2.5 Flash Native Audio Dialog Thinking0.91
  15. 15GPT-Live-1 (Astra, medium)0.9
  16. 16GPT-Live-1 (Sol, low)0.89
  17. 17Nova 2.0 Sonic0.88
  18. 18GPT-Realtime-2.1 Minimal0.87
  19. 19Deepslate Opal0.85
  20. 20GPT Realtime (Aug 2025)0.83
  21. 21GPT-Realtime-1.50.81
  22. 22GPT-Realtime-2.1 Mini High0.75
  23. 23GPT-Realtime-2 Minimal0.72
  24. 24Gemini 3.1 Flash Audio Preview (Mar 2026) - Minimal0.71
  25. 25Gemini 2.5 Flash Native Audio Dialog0.69
  26. 26GPT-4o mini Realtime (Dec '24)0.69
  27. 27Higgs Realtime0.69
  28. 28GPT Realtime Mini (Oct 2025)0.64
  29. 29GPT-Realtime-2.1 Mini Minimal0.63
  30. 30Qwen3.5 Omni Flash Realtime0.59
  31. 31Qwen3 Omni 30B A3B Instruct0.59
  32. 32Raon SpeechChat0.57
  33. 33Qwen3 Omni 30B A3B Instruct (Realtime)0.57
  34. 34GPT 4o audio chatcompletions0.54
  35. 35Freeze-Omni0.33
  36. 36Nemotron 3 VoiceChat (V1)0.27
  37. 37PersonaPlex0.19
  38. 38FLM-Audio0.16
  39. 39Moshi0.04
  40. 40Gemini 3.8 Live Extended Thinking (High)—
  41. 41GPT-5 (ChatGPT)—
  42. 42Gemini 2.5 Flash Native Audio Preview (Sep 2025)—
  43. 43Gemini 2.5 Flash Native Audio Preview (Dec 2025)—
  44. 44GPT-4o Realtime (Dec 2024)—

语音转文本

词错误率指数 · 71
  1. 1GPT-4o Mini Transcribe, OpenAI0
  2. 2Speechmatics Enhanced0
  3. 3Voxtral Small, Mistral0
  4. 4Solaria-1, Gladia0
  5. 5Smallest AI Pulse Pro0
  6. 6Gemini 3 Flash (High), Google0
  7. 7Nova 2 Pro, Amazon0
  8. 8StepAudio 2.5 ASR, StepFun0
  9. 9GPT-4o Transcribe, OpenAI0
  10. 10GPT Transcribe, OpenAI0
  11. 11Grok Voice Transcribe 2.00
  12. 12Gemini 2.0 Flash Lite, Google0
  13. 13Gemini 3.1 Pro Preview (High)0
  14. 14Fun-Realtime-ASR-preview0
  15. 15Inworld STT 10
  16. 16MAI-Transcribe-10
  17. 17Chirp 3, Google0
  18. 18Whisper Large v2, OpenAI0
  19. 19Universal, AssemblyAI0
  20. 20Gemini 3.1 Flash-Lite Preview (Minimal)0
  21. 21Qwen3.5 Omni Plus0
  22. 22Soniox V40
  23. 23Gemini 2.5 Pro, Google0
  24. 24Gemini 3.1 Pro Preview (Low)0
  25. 25Voxtral Mini Transcribe 2, Mistral0
  26. 26Scribe v2, ElevenLabs0
  27. 27Universal-3 Pro, AssemblyAI0
  28. 28Whisper Large v3 Turbo, OpenAI0
  29. 29Cohere Transcribe0
  30. 30Smallest AI Pulse0
  31. 31MAI-Transcribe-1.50
  32. 32Grok Voice Transcribe 1.00
  33. 33LLM Speech, Azure0
  34. 34Modulate STT Batch English VFast0
  35. 35Universal-3.5 Pro, AssemblyAI0
  36. 36Melia0
  37. 37Resonant-10
  38. 38Canary Qwen 2.5B, NVIDIA0
  39. 39Solaria-3, Gladia0
  40. 40Amazon Transcribe0
  41. 41Gemini 3.5 Transcribe0
  42. 42Inkling (256K), Thinking Machines0
  43. 43Soniox v5 Async0
  44. 44Whisper Large v3, OpenAI0
  45. 45StepAudio 3 ASR0
  46. 46MAI-Transcribe-20
  47. 47Gemini 2.5 Flash, Google0.1
  48. 48Nova 2 Omni, Amazon0.1
  49. 49Gemini 2.5 Lite, Google0.1
  50. 50Speechmatics Standard0.1
  51. 51Parakeet RNNT 1.1B, NVIDIA0.1
  52. 52Parakeet TDT 0.6B V2, NVIDIA0.1
  53. 53Raon Speech0.1
  54. 54Gemma 4 12B, Google0.1
  55. 55Gradium Speech-to-Text0.1
  56. 56Nova-3, Deepgram0.1
  57. 57Qwen3.5 Omni Flash0.1
  58. 58Chirp 2, Google0.1
  59. 59Rev AI0.1
  60. 60Qwen3 ASR Flash, Alibaba0.1
  61. 61Cloud Speech-To-Text (Chirp), Google0.3
  62. 62WhisperX—
  63. 63Qwen3 ASR 1.7B—
  64. 64Gemini 3.1 Flash-Lite Preview (High)—
  65. 65Qwen3 ASR 0.6B—
  66. 66Qwen3 Omni, Alibaba—
  67. 67Gemini 3 Flash (Minimal), Google—
  68. 68Gemini 2.5 Pro (Low Budget Reasoning), Google—
  69. 69GPT Live Transcribe, OpenAI—
  70. 70Distil-Whisper Large v2, Hugging Face—
  71. 71Modulate STT Streaming English—

有 46 个模型在已发布的精度下并列最优,这份榜单区分不了它们的先后。

越低越好

纯音乐

Elo · 20
  1. 1Mureka V91,177±17
  2. 2Suno V5.51,171±15
  3. 3Mureka V81,143±13
  4. 4Suno V51,137±14
  5. 5StepAudio 3 Music1,113±14
  6. 6Lyria 3 Pro1,102±13
  7. 7Lyria 3.51,085±14
  8. 8Suno V4.51,065±20
  9. 9MiniMax Music 2.61,063±13
  10. 10Eleven Music v21,050±14
  11. 11MiniMax Music 2.5+1,035±14
  12. 12FUZZ-2.0 Raw1,034±22
  13. 13FUZZ-1.1 Pro1,014±22
  14. 14Eleven Music1,002±22
  15. 15Lyria 21,000±0
  16. 16FUZZ-2.0998±23
  17. 17Stable Audio 2.0958±27
  18. 18Sonauto V2.1937±27
  19. 19Udio v1.5 Allegro920±28
  20. 20MusicGen846±34

Elo 846–1177

误差条重叠的模型,排名先后没有统计意义

带人声音乐

Elo · 17
  1. 1Suno V5.51,160±14
  2. 2Mureka V91,136±16
  3. 3Mureka V81,129±13
  4. 4StepAudio 3 Music1,092±15
  5. 5MiniMax Music 2.5+1,084±13
  6. 6Suno V51,077±13
  7. 7Lyria 3 Pro1,074±13
  8. 8MiniMax Music 2.61,071±13
  9. 9Lyria 3.51,067±15
  10. 10Suno V4.51,062±18
  11. 11FUZZ-2.01,051±18
  12. 12Eleven Music v21,047±14
  13. 13Eleven Music1,022±19
  14. 14FUZZ-1.1 Pro1,000±0
  15. 15FUZZ-2.0 Raw1,000±19
  16. 16Sonauto V2.1972±22
  17. 17Udio v1.5 Allegro956±22

Elo 956–1160

误差条重叠的模型,排名先后没有统计意义