Benchmark Almanac

Razavi-Bench

Gemini 4 Carbon 98.3% leads since October 2026
Not in model cards

Models explain what happens in analog circuits and why. A language model awards up to four points per answer against a reference and a rubric.

The authors rank 73 models on 50 questions. Each model answers every question three times without tools.

Kind
Electronics
Level
University
Scored
Graded by a model
Answers
Open-ended
Tasks
All public
Measures
Capability
Language
English
In and out
Text, Images → Text
Interaction
Single turn
Items from
Written by experts
Refresh
One fixed set
Run by
Anyone
In views
Vision
Strongest model tested
Claude Fable 5.2 #2 on the capability index
Last model added
November 2026
Scores refreshed
November 20th, 2026
73
Entries on its own leaderboard, 63 models we track

Gemini 4 Carbon leads with 98.3% Gemini 4 Carbon leads with 98.3%

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.

  1. 1 Gemini 4 Carbon high 98.3%
  2. 2 Claude Opus 5.5 max 97.7%
  3. 3 Claude Fable 5.2 max 95.8%
  4. 4 Gemini 4 Argon high 94.8%
  5. 5 GPT-6 Astra max 94.7%
  6. 6 Claude Opus 5 max 94%
  7. 7 Claude Fable 5.1 max 93.7%
  8. 8 GPT-6.1 Sol max 92.8%
  9. 9 Claude Fable 5 max 90.5%
  10. 10 Tencent Hy4 high 90.3%
  11. 11 GPT-5.6 Sol max 89.7%
  12. 12 Gemini 3.8 Flash high 88.7%
  13. 13 Gemini 3.7 Flash high 88.5%
  14. 14 Muse Wonder Full 1.0 max 87.8%
  15. 15 GPT-6 Sol max 87.8%
  16. 16 GPT-5.6 Terra max 86.7%
  17. 17 GPT-5.5 xhigh 86.5%
  18. 18 Gemini 3.5 Flash high 85.5%
  19. 19 Kimi 3.1 Pro max 84.8%
  20. 20 Gemini 3.6 Flash high 84%
  21. 21 Muse Spark 1.2 xhigh 83.7%
  22. 22 GPT-5.4 xhigh 83%
  23. 23 Claude Opus 4.8 max 82.7%
  24. 24 GPT-5.3 xhigh 82.3%
  25. 25 Qwen 3.8 Max xhigh 81.3%
  26. 26 Muse Spark 1.3 max 80.5%
  27. 27 Claude Haiku 5.5 max 80.2%
  28. 28 GLM-5.3-Flash max 79.8%
  29. 29 Gemini 3 Flash high 79.3%
  30. 30 Gemini 3.1 Pro high 77.7%
  31. 31 DeepSeek V4.1 Flash 1017 max 77.3%
  32. 32 GPT-5.6 Luna max 77%
  33. 33 Grok 4.6 xhigh 75.8%
  34. 34 GPT-6 Luna max 75.5%
  35. 35 Grok 4.5 high 74.5%
  36. 36 GPT-5.2 xhigh 73.7%
  37. 37 Kimi K3 max 73.5%
  38. 38 MiniMax-M3.1-Flash max 72.7%
  39. 39 Qwen3.7-Plus max 71.8%
  40. 40 Grok 4.7 xhigh 71.2%
  41. 41 Mistral Large 4 high 70.3%
  42. 42 DeepSeek V4.1 Flash 0910 max 69.7%
  43. 43 MiniMax-M3 thinking 69.2%
  44. 44 Qwen 3.8 27B xhigh 67%
  45. 45 GPT-5.4 Mini xhigh 66.8%
  46. 46 Qwen3.7 Flash max 65%
  47. 47 Kimi K2.7 Code thinking 64.2%
  48. 48 Claude Sonnet 4.6 max 62.2%
  49. 49 Inkling-Small max 58.5%
  50. 50 Gemini 2.5 Pro Jun 2025, max 56.8%
  51. 51 GPT-5.1 high 55%
  52. 52 Mimo 2.6 Pro max 54.8%
  53. 53 GPT-5 high 53.3%
  54. 54 Claude Opus 4.5 max 52.8%
  55. 55 o3 high 51.5%
  56. 56 Gemma 4 31B IT max 51.5%
  57. 57 Inkling max 48.7%
  58. 58 Step 3.7 Flash high 47.5%
  59. 59 Llama 4 Maverick default 46%
  60. 60 Claude Sonnet 4.5 max 44.2%
  61. 61 Qwen3.5-9B thinking 43.2%
  62. 62 Gemini 2.5 Flash-Lite Sep 2025, max 42.5%
  63. 63 Nemotron 3.5 Super VL max 37.5%
  64. 64 Mistral Medium 3.5 high 36.5%
  65. 65 Qwen3-VL-235B-A22B-Instruct default 36%
  66. 66 GPT-4o (2024-11-20) default 31.7%
  67. 67 GPT-4o (2024-08-06) default 31.2%
  68. 68 Claude Haiku 4.5 max 31.2%
  69. 69 GPT-4o (2024-05-13) default 28.8%
  70. 70 Mistral Large 2512 default 28%
  71. 71 Qwen2.5-VL-72B default 26.3%
  72. 72 GPT-4o mini default 25.7%
  73. 73 Nova Pro default 21.7%

Source: Official Razavi-Bench leaderboard

  1. 1 Gemini 4 Carbon high 98.3%
  2. 2 Claude Opus 5.5 max 97.7%
  3. 3 Claude Fable 5.2 max 95.8%
  4. 4 Gemini 4 Argon high 94.8%
  5. 5 GPT-6 Astra max 94.7%
  6. 6 Claude Opus 5 max 94%
  7. 7 Claude Fable 5.1 max 93.7%
  8. 8 GPT-6.1 Sol max 92.8%
  9. 9 Claude Fable 5 max 90.5%
  10. 10 Tencent Hy4 high 90.3%
  11. 11 GPT-5.6 Sol max 89.7%
  12. 12 Gemini 3.8 Flash high 88.7%
  13. 13 Gemini 3.7 Flash high 88.5%
  14. 14 GPT-6 Sol max 87.8%
  15. 15 GPT-5.6 Terra max 86.7%
  16. 16 GPT-5.5 xhigh 86.5%
  17. 17 Gemini 3.5 Flash high 85.5%
  18. 18 Gemini 3.6 Flash high 84%
  19. 19 Muse Spark 1.2 xhigh 83.7%
  20. 20 GPT-5.4 xhigh 83%
  21. 21 Claude Opus 4.8 max 82.7%
  22. 22 Qwen 3.8 Max xhigh 81.3%
  23. 23 Muse Spark 1.3 max 80.5%
  24. 24 Claude Haiku 5.5 max 80.2%
  25. 25 GLM-5.3-Flash max 79.8%
  26. 26 Gemini 3 Flash high 79.3%
  27. 27 Gemini 3.1 Pro high 77.7%
  28. 28 DeepSeek V4.1 Flash 1017 max 77.3%
  29. 29 GPT-5.6 Luna max 77%
  30. 30 Grok 4.6 xhigh 75.8%
  31. 31 GPT-6 Luna max 75.5%
  32. 32 Grok 4.5 high 74.5%
  33. 33 GPT-5.2 xhigh 73.7%
  34. 34 Kimi K3 max 73.5%
  35. 35 MiniMax-M3.1-Flash max 72.7%
  36. 36 Qwen3.7-Plus max 71.8%
  37. 37 Grok 4.7 xhigh 71.2%
  38. 38 Mistral Large 4 high 70.3%
  39. 39 MiniMax-M3 thinking 69.2%
  40. 40 Qwen 3.8 27B xhigh 67%
  41. 41 GPT-5.4 Mini xhigh 66.8%
  42. 42 Qwen3.7 Flash max 65%
  43. 43 Kimi K2.7 Code thinking 64.2%
  44. 44 Claude Sonnet 4.6 max 62.2%
  45. 45 Inkling-Small max 58.5%
  46. 46 Gemini 2.5 Pro Jun 2025, max 56.8%
  47. 47 GPT-5.1 high 55%
  48. 48 GPT-5 high 53.3%
  49. 49 Claude Opus 4.5 max 52.8%
  50. 50 o3 high 51.5%
  51. 51 Gemma 4 31B IT max 51.5%
  52. 52 Inkling max 48.7%
  53. 53 Step 3.7 Flash high 47.5%
  54. 54 Llama 4 Maverick default 46%
  55. 55 Claude Sonnet 4.5 max 44.2%
  56. 56 Qwen3.5-9B thinking 43.2%
  57. 57 Gemini 2.5 Flash-Lite Sep 2025, max 42.5%
  58. 58 Nemotron 3.5 Super VL max 37.5%
  59. 59 Mistral Medium 3.5 high 36.5%
  60. 60 Qwen3-VL-235B-A22B-Instruct default 36%
  61. 61 Claude Haiku 4.5 max 31.2%
  62. 62 Qwen2.5-VL-72B default 26.3%
  63. 63 GPT-4o mini default 25.7%

Source: Official Razavi-Bench leaderboard

#EntryScore ±SettingDateLinkSource
1 Gemini 4 Carbon 98.3% 98 to 98.5 high Oct 2026 razavi-bench.tokenzhang.com Official leaderboard
2 Claude Opus 5.5 97.7% 97 to 98.5 max Sep 2026 razavi-bench.tokenzhang.com Official leaderboard
3 Claude Fable 5.2 95.8% 95 to 96.5 max Nov 2026 razavi-bench.tokenzhang.com Official leaderboard
4 Gemini 4 Argon 94.8% 93.5 to 95.5 high Sep 2026 razavi-bench.tokenzhang.com Official leaderboard
5 GPT-6 Astra 94.7% 94 to 95.5 max Sep 2026 razavi-bench.tokenzhang.com Official leaderboard
6 Claude Opus 5 94% 92.5 to 95 max Jul 2026 razavi-bench.tokenzhang.com Official leaderboard
7 Claude Fable 5.1 93.7% 92.5 to 95 max Sep 2026 razavi-bench.tokenzhang.com Official leaderboard
8 GPT-6.1 Sol 92.8% 91 to 95 max Sep 2026 razavi-bench.tokenzhang.com Official leaderboard
9 Claude Fable 5 90.5% ±0.5 max Jun 2026 razavi-bench.tokenzhang.com Official leaderboard
10 Tencent Hy4 90.3% 88 to 91.5 high Nov 2026 razavi-bench.tokenzhang.com Official leaderboard
11 GPT-5.6 Sol 89.7% 88.5 to 90.5 max Jul 2026 razavi-bench.tokenzhang.com Official leaderboard
12 Gemini 3.8 Flash 88.7% 88.5 to 89 high Sep 2026 razavi-bench.tokenzhang.com Official leaderboard
13 Gemini 3.7 Flash 88.5% ±1 high Aug 2026 razavi-bench.tokenzhang.com Official leaderboard
14 Muse Wonder Full 1.0 87.8% 84.5 to 90.5 max razavi-bench.tokenzhang.com Official leaderboard
15 GPT-6 Sol 87.8% 86 to 89 max Sep 2026 razavi-bench.tokenzhang.com Official leaderboard
16 GPT-5.6 Terra 86.7% 85.5 to 87.5 max Jul 2026 razavi-bench.tokenzhang.com Official leaderboard
17 GPT-5.5 86.5% 85 to 87.5 xhigh Apr 2026 razavi-bench.tokenzhang.com Official leaderboard
18 Gemini 3.5 Flash 85.5% 84.5 to 87 high May 2026 razavi-bench.tokenzhang.com Official leaderboard
19 Kimi 3.1 Pro 84.8% 84.5 to 85 max razavi-bench.tokenzhang.com Official leaderboard
20 Gemini 3.6 Flash 84% ±0.5 high Jul 2026 razavi-bench.tokenzhang.com Official leaderboard
21 Muse Spark 1.2 83.7% 81.5 to 85 xhigh Aug 2026 razavi-bench.tokenzhang.com Official leaderboard
22 GPT-5.4 83% ±0.5 xhigh Mar 2026 razavi-bench.tokenzhang.com Official leaderboard
23 Claude Opus 4.8 82.7% 77 to 85.5 max May 2026 razavi-bench.tokenzhang.com Official leaderboard
24 GPT-5.3 82.3% 81.5 to 83 xhigh razavi-bench.tokenzhang.com Official leaderboard
25 Qwen 3.8 Max 81.3% 80.5 to 82 xhigh Aug 2026 razavi-bench.tokenzhang.com Official leaderboard
26 Muse Spark 1.3 80.5% 79 to 81.5 max Sep 2026 razavi-bench.tokenzhang.com Official leaderboard
27 Claude Haiku 5.5 80.2% 79.5 to 80.5 max Oct 2026 razavi-bench.tokenzhang.com Official leaderboard
28 GLM-5.3-Flash 79.8% 79.5 to 80 max Aug 2026 razavi-bench.tokenzhang.com Official leaderboard
29 Gemini 3 Flash 79.3% 76.5 to 81.5 high Dec 2025 razavi-bench.tokenzhang.com Official leaderboard
30 Gemini 3.1 Pro 77.7% 76.5 to 78.5 high Feb 2026 razavi-bench.tokenzhang.com Official leaderboard
31 DeepSeek V4.1 Flash 1017 77.3% ±4.3 max Oct 2026 razavi-bench.tokenzhang.com Official leaderboard
32 GPT-5.6 Luna 77% 75 to 78.5 max Jul 2026 razavi-bench.tokenzhang.com Official leaderboard
33 Grok 4.6 75.8% 74 to 78.5 xhigh Aug 2026 razavi-bench.tokenzhang.com Official leaderboard
34 GPT-6 Luna 75.5% 75 to 76.5 max Sep 2026 razavi-bench.tokenzhang.com Official leaderboard
35 Grok 4.5 74.5% ±1.5 high Jul 2026 razavi-bench.tokenzhang.com Official leaderboard
36 GPT-5.2 73.7% 71.5 to 75 xhigh Dec 2025 razavi-bench.tokenzhang.com Official leaderboard
37 Kimi K3 73.5% 70.5 to 76 max Jul 2026 razavi-bench.tokenzhang.com Official leaderboard
38 MiniMax-M3.1-Flash 72.7% 70 to 78 max Oct 2026 razavi-bench.tokenzhang.com Official leaderboard
39 Qwen3.7-Plus 71.8% 69 to 74 max Jun 2026 razavi-bench.tokenzhang.com Official leaderboard
40 Grok 4.7 71.2% 70 to 72 xhigh Sep 2026 razavi-bench.tokenzhang.com Official leaderboard
41 Mistral Large 4 70.3% 68.5 to 71.5 high Oct 2026 razavi-bench.tokenzhang.com Official leaderboard
42 DeepSeek V4.1 Flash 0910 69.7% 68.5 to 71 max razavi-bench.tokenzhang.com Official leaderboard
43 MiniMax-M3 69.2% 68 to 70.5 thinking Jun 2026 razavi-bench.tokenzhang.com Official leaderboard
44 Qwen 3.8 27B 67% 64.5 to 68.5 xhigh Aug 2026 razavi-bench.tokenzhang.com Official leaderboard
45 GPT-5.4 Mini 66.8% 64.5 to 69.5 xhigh Mar 2026 razavi-bench.tokenzhang.com Official leaderboard
46 Qwen3.7 Flash 65% 63.5 to 66 max Jul 2026 razavi-bench.tokenzhang.com Official leaderboard
47 Kimi K2.7 Code 64.2% 60.5 to 66.5 thinking Jun 2026 razavi-bench.tokenzhang.com Official leaderboard
48 Claude Sonnet 4.6 62.2% 61.5 to 63 max Feb 2026 razavi-bench.tokenzhang.com Official leaderboard
49 Inkling-Small 58.5% 56.5 to 62 max Jul 2026 razavi-bench.tokenzhang.com Official leaderboard
50 Gemini 2.5 Pro Jun 2025 56.8% 56 to 57.5 max Jun 2025 razavi-bench.tokenzhang.com Official leaderboard
51 GPT-5.1 55% 52.5 to 59.5 high Nov 2025 razavi-bench.tokenzhang.com Official leaderboard
52 Mimo 2.6 Pro 54.8% 53 to 57 max razavi-bench.tokenzhang.com Official leaderboard
53 GPT-5 53.3% 51.5 to 55 high Aug 2025 razavi-bench.tokenzhang.com Official leaderboard
54 Claude Opus 4.5 52.8% 50 to 54.5 max Nov 2025 razavi-bench.tokenzhang.com Official leaderboard
55 o3 51.5% ±1.5 high Apr 2025 razavi-bench.tokenzhang.com Official leaderboard
56 Gemma 4 31B IT 51.5% 47 to 55 max Apr 2026 razavi-bench.tokenzhang.com Official leaderboard
57 Inkling 48.7% 42.5 to 54 max Jul 2026 razavi-bench.tokenzhang.com Official leaderboard
58 Step 3.7 Flash 47.5% 45 to 50.5 high May 2026 razavi-bench.tokenzhang.com Official leaderboard
59 Llama 4 Maverick 46% 44.5 to 48.5 default Apr 2025 razavi-bench.tokenzhang.com Official leaderboard
60 Claude Sonnet 4.5 44.2% 42.5 to 46.5 max Sep 2025 razavi-bench.tokenzhang.com Official leaderboard
61 Qwen3.5-9B 43.2% 42 to 44 thinking Feb 2026 razavi-bench.tokenzhang.com Official leaderboard
62 Gemini 2.5 Flash-Lite Sep 2025 42.5% max Sep 2025 razavi-bench.tokenzhang.com Official leaderboard
63 Nemotron 3.5 Super VL 37.5% 35 to 39.5 max Oct 2026 razavi-bench.tokenzhang.com Official leaderboard
64 Mistral Medium 3.5 36.5% 33.5 to 40 high Apr 2026 razavi-bench.tokenzhang.com Official leaderboard
65 Qwen3-VL-235B-A22B-Instruct 36% 33 to 38 default Sep 2025 razavi-bench.tokenzhang.com Official leaderboard
66 GPT-4o (2024-11-20) 31.7% 30.5 to 32.5 default razavi-bench.tokenzhang.com Official leaderboard
67 GPT-4o (2024-08-06) 31.2% 30 to 32 default razavi-bench.tokenzhang.com Official leaderboard
68 Claude Haiku 4.5 31.2% 27.5 to 33.5 max Oct 2025 razavi-bench.tokenzhang.com Official leaderboard
69 GPT-4o (2024-05-13) 28.8% 28.5 to 29.5 default razavi-bench.tokenzhang.com Official leaderboard
70 Mistral Large 2512 28% ±1 default razavi-bench.tokenzhang.com Official leaderboard
71 Qwen2.5-VL-72B 26.3% 25 to 27 default Jan 2025 razavi-bench.tokenzhang.com Official leaderboard
72 GPT-4o mini 25.7% 24.5 to 27 default Jul 2024 razavi-bench.tokenzhang.com Official leaderboard
73 Nova Pro 21.7% 20.5 to 23 default razavi-bench.tokenzhang.com Official leaderboard
#ModelScore ±SettingDateLinkSource
1 Gemini 4 Carbon 98.3% 98 to 98.5 high Oct 2026 razavi-bench.tokenzhang.com Official leaderboard
2 Claude Opus 5.5 97.7% 97 to 98.5 max Sep 2026 razavi-bench.tokenzhang.com Official leaderboard
3 Claude Fable 5.2 95.8% 95 to 96.5 max Nov 2026 razavi-bench.tokenzhang.com Official leaderboard
4 Gemini 4 Argon 94.8% 93.5 to 95.5 high Sep 2026 razavi-bench.tokenzhang.com Official leaderboard
5 GPT-6 Astra 94.7% 94 to 95.5 max Sep 2026 razavi-bench.tokenzhang.com Official leaderboard
6 Claude Opus 5 94% 92.5 to 95 max Jul 2026 razavi-bench.tokenzhang.com Official leaderboard
7 Claude Fable 5.1 93.7% 92.5 to 95 max Sep 2026 razavi-bench.tokenzhang.com Official leaderboard
8 GPT-6.1 Sol 92.8% 91 to 95 max Sep 2026 razavi-bench.tokenzhang.com Official leaderboard
9 Claude Fable 5 90.5% ±0.5 max Jun 2026 razavi-bench.tokenzhang.com Official leaderboard
10 Tencent Hy4 90.3% 88 to 91.5 high Nov 2026 razavi-bench.tokenzhang.com Official leaderboard
11 GPT-5.6 Sol 89.7% 88.5 to 90.5 max Jul 2026 razavi-bench.tokenzhang.com Official leaderboard
12 Gemini 3.8 Flash 88.7% 88.5 to 89 high Sep 2026 razavi-bench.tokenzhang.com Official leaderboard
13 Gemini 3.7 Flash 88.5% ±1 high Aug 2026 razavi-bench.tokenzhang.com Official leaderboard
14 GPT-6 Sol 87.8% 86 to 89 max Sep 2026 razavi-bench.tokenzhang.com Official leaderboard
15 GPT-5.6 Terra 86.7% 85.5 to 87.5 max Jul 2026 razavi-bench.tokenzhang.com Official leaderboard
16 GPT-5.5 86.5% 85 to 87.5 xhigh Apr 2026 razavi-bench.tokenzhang.com Official leaderboard
17 Gemini 3.5 Flash 85.5% 84.5 to 87 high May 2026 razavi-bench.tokenzhang.com Official leaderboard
18 Gemini 3.6 Flash 84% ±0.5 high Jul 2026 razavi-bench.tokenzhang.com Official leaderboard
19 Muse Spark 1.2 83.7% 81.5 to 85 xhigh Aug 2026 razavi-bench.tokenzhang.com Official leaderboard
20 GPT-5.4 83% ±0.5 xhigh Mar 2026 razavi-bench.tokenzhang.com Official leaderboard
21 Claude Opus 4.8 82.7% 77 to 85.5 max May 2026 razavi-bench.tokenzhang.com Official leaderboard
22 Qwen 3.8 Max 81.3% 80.5 to 82 xhigh Aug 2026 razavi-bench.tokenzhang.com Official leaderboard
23 Muse Spark 1.3 80.5% 79 to 81.5 max Sep 2026 razavi-bench.tokenzhang.com Official leaderboard
24 Claude Haiku 5.5 80.2% 79.5 to 80.5 max Oct 2026 razavi-bench.tokenzhang.com Official leaderboard
25 GLM-5.3-Flash 79.8% 79.5 to 80 max Aug 2026 razavi-bench.tokenzhang.com Official leaderboard
26 Gemini 3 Flash 79.3% 76.5 to 81.5 high Dec 2025 razavi-bench.tokenzhang.com Official leaderboard
27 Gemini 3.1 Pro 77.7% 76.5 to 78.5 high Feb 2026 razavi-bench.tokenzhang.com Official leaderboard
28 DeepSeek V4.1 Flash 1017 77.3% ±4.3 max Oct 2026 razavi-bench.tokenzhang.com Official leaderboard
29 GPT-5.6 Luna 77% 75 to 78.5 max Jul 2026 razavi-bench.tokenzhang.com Official leaderboard
30 Grok 4.6 75.8% 74 to 78.5 xhigh Aug 2026 razavi-bench.tokenzhang.com Official leaderboard
31 GPT-6 Luna 75.5% 75 to 76.5 max Sep 2026 razavi-bench.tokenzhang.com Official leaderboard
32 Grok 4.5 74.5% ±1.5 high Jul 2026 razavi-bench.tokenzhang.com Official leaderboard
33 GPT-5.2 73.7% 71.5 to 75 xhigh Dec 2025 razavi-bench.tokenzhang.com Official leaderboard
34 Kimi K3 73.5% 70.5 to 76 max Jul 2026 razavi-bench.tokenzhang.com Official leaderboard
35 MiniMax-M3.1-Flash 72.7% 70 to 78 max Oct 2026 razavi-bench.tokenzhang.com Official leaderboard
36 Qwen3.7-Plus 71.8% 69 to 74 max Jun 2026 razavi-bench.tokenzhang.com Official leaderboard
37 Grok 4.7 71.2% 70 to 72 xhigh Sep 2026 razavi-bench.tokenzhang.com Official leaderboard
38 Mistral Large 4 70.3% 68.5 to 71.5 high Oct 2026 razavi-bench.tokenzhang.com Official leaderboard
39 MiniMax-M3 69.2% 68 to 70.5 thinking Jun 2026 razavi-bench.tokenzhang.com Official leaderboard
40 Qwen 3.8 27B 67% 64.5 to 68.5 xhigh Aug 2026 razavi-bench.tokenzhang.com Official leaderboard
41 GPT-5.4 Mini 66.8% 64.5 to 69.5 xhigh Mar 2026 razavi-bench.tokenzhang.com Official leaderboard
42 Qwen3.7 Flash 65% 63.5 to 66 max Jul 2026 razavi-bench.tokenzhang.com Official leaderboard
43 Kimi K2.7 Code 64.2% 60.5 to 66.5 thinking Jun 2026 razavi-bench.tokenzhang.com Official leaderboard
44 Claude Sonnet 4.6 62.2% 61.5 to 63 max Feb 2026 razavi-bench.tokenzhang.com Official leaderboard
45 Inkling-Small 58.5% 56.5 to 62 max Jul 2026 razavi-bench.tokenzhang.com Official leaderboard
46 Gemini 2.5 Pro Jun 2025 56.8% 56 to 57.5 max Jun 2025 razavi-bench.tokenzhang.com Official leaderboard
47 GPT-5.1 55% 52.5 to 59.5 high Nov 2025 razavi-bench.tokenzhang.com Official leaderboard
48 GPT-5 53.3% 51.5 to 55 high Aug 2025 razavi-bench.tokenzhang.com Official leaderboard
49 Claude Opus 4.5 52.8% 50 to 54.5 max Nov 2025 razavi-bench.tokenzhang.com Official leaderboard
50 o3 51.5% ±1.5 high Apr 2025 razavi-bench.tokenzhang.com Official leaderboard
51 Gemma 4 31B IT 51.5% 47 to 55 max Apr 2026 razavi-bench.tokenzhang.com Official leaderboard
52 Inkling 48.7% 42.5 to 54 max Jul 2026 razavi-bench.tokenzhang.com Official leaderboard
53 Step 3.7 Flash 47.5% 45 to 50.5 high May 2026 razavi-bench.tokenzhang.com Official leaderboard
54 Llama 4 Maverick 46% 44.5 to 48.5 default Apr 2025 razavi-bench.tokenzhang.com Official leaderboard
55 Claude Sonnet 4.5 44.2% 42.5 to 46.5 max Sep 2025 razavi-bench.tokenzhang.com Official leaderboard
56 Qwen3.5-9B 43.2% 42 to 44 thinking Feb 2026 razavi-bench.tokenzhang.com Official leaderboard
57 Gemini 2.5 Flash-Lite Sep 2025 42.5% max Sep 2025 razavi-bench.tokenzhang.com Official leaderboard
58 Nemotron 3.5 Super VL 37.5% 35 to 39.5 max Oct 2026 razavi-bench.tokenzhang.com Official leaderboard
59 Mistral Medium 3.5 36.5% 33.5 to 40 high Apr 2026 razavi-bench.tokenzhang.com Official leaderboard
60 Qwen3-VL-235B-A22B-Instruct 36% 33 to 38 default Sep 2025 razavi-bench.tokenzhang.com Official leaderboard
61 Claude Haiku 4.5 31.2% 27.5 to 33.5 max Oct 2025 razavi-bench.tokenzhang.com Official leaderboard
62 Qwen2.5-VL-72B 26.3% 25 to 27 default Jan 2025 razavi-bench.tokenzhang.com Official leaderboard
63 GPT-4o mini 25.7% 24.5 to 27 default Jul 2024 razavi-bench.tokenzhang.com Official leaderboard

Rankings by part

We read 2 parts from the benchmark’s own leaderboard: Part 1 and Part 2. Each ranks models on one part of the test. Claude Opus 5.5 leads Part 1 and Gemini 4 Carbon leads Part 2. Arrows show how far a model moved from its overall rank among the models we track.

Part 1

63 models we track have a score here.

  1. 1 Claude Opus 5.5 ↑1 100%
  2. 2 Claude Fable 5.2 ↑1 100%
  3. 3 Claude Fable 5.1 ↑4 99.7%
  4. 4 GPT-6 Astra ↑1 99.7%
  5. 5 Gemini 4 Carbon ↓4 99.7%
  6. 6 GPT-6.1 Sol ↑2 99.4%
  7. 7 Claude Fable 5 ↑2 98.9%
  8. 8 Gemini 3.8 Flash ↑4 98.6%
  9. 9 Gemini 4 Argon ↓5 98.3%
  10. 10 Gemini 3.7 Flash ↑3 98.1%

The best score, 98.3%, costs $28 for the full run

Runs on the cost frontier score higher than any run that's cheaper. Gemini 4 Argon at high comes within 3.5 points of the best for $8.19 for the full run, 3 times cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.

Meta Google DeepMind Qwen OpenAI DeepSeek Tencent Cost frontier One model at each effort level
0% 20% 40% 60% 80% 100% $0.10 $1.00 $10 $100 Cost of the full run, log scale Gemini 4 Carbon, high Gemini 4 Argon, high Tencent Hy4, high Gemini 3 Flash, high GPT-6 Luna, max Qwen3.7 Flash, max Llama 4 Maverick, default 0% 20% 40% 60% 80% 100% $0.10 $1.00 $10 $100 Cost of the full run, log scale Gemini 4 Carbon, high Tencent Hy4, high

Source: Official Razavi-Bench leaderboard

ModelEffortHarnessScoreCost of the full runOutput tokens
Gemini 4 Carbon high 98.3% $28
Claude Opus 5.5 max 97.7% $68
Claude Fable 5.2 max 95.8% $51
Gemini 4 Argon high 94.8% $8.19
GPT-6 Astra max 94.7% $43
Claude Opus 5 max 94% $28
Claude Fable 5.1 max 93.7% $93
GPT-6.1 Sol max 92.8% $8.34
Claude Fable 5 max 90.5% $79
Tencent Hy4 high 90.3% $4.77
GPT-5.6 Sol max 89.7% $35
Gemini 3.8 Flash high 88.7% $7.28
Gemini 3.7 Flash high 88.5% $2.73
GPT-6 Sol max 87.8% $8.14
GPT-5.6 Terra max 86.7% $35
GPT-5.5 xhigh 86.5% $39
Gemini 3.5 Flash high 85.5% $7.94
Gemini 3.6 Flash high 84% $2.95
Muse Spark 1.2 xhigh 83.7% $6.61
GPT-5.4 xhigh 83% $32
Claude Opus 4.8 max 82.7% $76
Qwen 3.8 Max xhigh 81.3% $9.15
Muse Spark 1.3 max 80.5% $7.57
Claude Haiku 5.5 max 80.2% $4.68
GLM-5.3-Flash max 79.8% $2.74
Gemini 3 Flash high 79.3% $2.41
Gemini 3.1 Pro high 77.7% $23
DeepSeek V4.1 Flash 1017 max 77.3% $1.92
GPT-5.6 Luna max 77% $3.36
Grok 4.6 xhigh 75.8% $5.81
GPT-6 Luna max 75.5% $0.93
Grok 4.5 high 74.5% $4.55
GPT-5.2 xhigh 73.7% $39
Kimi K3 max 73.5% $34
MiniMax-M3.1-Flash max 72.7% $1.62
Qwen3.7-Plus max 71.8% $1.65
Grok 4.7 xhigh 71.2% $13
Mistral Large 4 high 70.3% $6.31
MiniMax-M3 thinking 69.2% $3.46
Qwen 3.8 27B xhigh 67% $3.63
GPT-5.4 Mini xhigh 66.8% $13
Qwen3.7 Flash max 65% $0.22
Kimi K2.7 Code thinking 64.2% $5.60
Claude Sonnet 4.6 max 62.2% $68
Inkling-Small max 58.5% $4.02
Gemini 2.5 Pro Jun 2025 max 56.8% $8.22
GPT-5.1 high 55% $14
GPT-5 high 53.3% $13
Claude Opus 4.5 max 52.8% $20
o3 high 51.5% $5.90
Gemma 4 31B IT max 51.5% $0.23
Inkling max 48.7% $11
Step 3.7 Flash high 47.5% $17
Llama 4 Maverick default 46% $0.18
Claude Sonnet 4.5 max 44.2% $7.54
Qwen3.5-9B thinking 43.2% $0.70
Gemini 2.5 Flash-Lite Sep 2025 max 42.5% $2.72
Nemotron 3.5 Super VL max 37.5% $1.26
Mistral Medium 3.5 high 36.5% $7.77
Qwen3-VL-235B-A22B-Instruct default 36% $0.39
Claude Haiku 4.5 max 31.2% $3.02
Qwen2.5-VL-72B default 26.3% $0.31
GPT-4o mini default 25.7% $0.56

The best score, 98.3%, costs $28 for the full run

Runs on the cost frontier score higher than any run that's cheaper. Gemini 4 Argon at high comes within 3.5 points of the best for $8.19 for the full run, 3 times cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.

Meta Google DeepMind Qwen OpenAI DeepSeek Tencent Cost frontier One model at each effort level
0% 20% 40% 60% 80% 100% $0.10 $1.00 $10 $100 Cost of the full run, log scale Gemini 4 Carbon, high Gemini 4 Argon, high Tencent Hy4, high Gemini 3 Flash, high GPT-6 Luna, max Qwen3.7 Flash, max Llama 4 Maverick, default 0% 20% 40% 60% 80% 100% $0.10 $1.00 $10 $100 Cost of the full run, log scale Gemini 4 Carbon, high Tencent Hy4, high

Source: Official Razavi-Bench leaderboard

ModelEffortHarnessScoreCost of the full runOutput tokens
Gemini 4 Carbon high 98.3% $28
Claude Opus 5.5 max 97.7% $68
Claude Fable 5.2 max 95.8% $51
Gemini 4 Argon high 94.8% $8.19
GPT-6 Astra max 94.7% $43
Claude Opus 5 max 94% $28
Claude Fable 5.1 max 93.7% $93
GPT-6.1 Sol max 92.8% $8.34
Claude Fable 5 max 90.5% $79
Tencent Hy4 high 90.3% $4.77
GPT-5.6 Sol max 89.7% $35
Gemini 3.8 Flash high 88.7% $7.28
Gemini 3.7 Flash high 88.5% $2.73
GPT-6 Sol max 87.8% $8.14
GPT-5.6 Terra max 86.7% $35
GPT-5.5 xhigh 86.5% $39
Gemini 3.5 Flash high 85.5% $7.94
Gemini 3.6 Flash high 84% $2.95
Muse Spark 1.2 xhigh 83.7% $6.61
GPT-5.4 xhigh 83% $32
Claude Opus 4.8 max 82.7% $76
Qwen 3.8 Max xhigh 81.3% $9.15
Muse Spark 1.3 max 80.5% $7.57
Claude Haiku 5.5 max 80.2% $4.68
GLM-5.3-Flash max 79.8% $2.74
Gemini 3 Flash high 79.3% $2.41
Gemini 3.1 Pro high 77.7% $23
DeepSeek V4.1 Flash 1017 max 77.3% $1.92
GPT-5.6 Luna max 77% $3.36
Grok 4.6 xhigh 75.8% $5.81
GPT-6 Luna max 75.5% $0.93
Grok 4.5 high 74.5% $4.55
GPT-5.2 xhigh 73.7% $39
Kimi K3 max 73.5% $34
MiniMax-M3.1-Flash max 72.7% $1.62
Qwen3.7-Plus max 71.8% $1.65
Grok 4.7 xhigh 71.2% $13
Mistral Large 4 high 70.3% $6.31
MiniMax-M3 thinking 69.2% $3.46
Qwen 3.8 27B xhigh 67% $3.63
GPT-5.4 Mini xhigh 66.8% $13
Qwen3.7 Flash max 65% $0.22
Kimi K2.7 Code thinking 64.2% $5.60
Claude Sonnet 4.6 max 62.2% $68
Inkling-Small max 58.5% $4.02
Gemini 2.5 Pro Jun 2025 max 56.8% $8.22
GPT-5.1 high 55% $14
GPT-5 high 53.3% $13
Claude Opus 4.5 max 52.8% $20
o3 high 51.5% $5.90
Gemma 4 31B IT max 51.5% $0.23
Inkling max 48.7% $11
Step 3.7 Flash high 47.5% $17
Llama 4 Maverick default 46% $0.18
Claude Sonnet 4.5 max 44.2% $7.54
Qwen3.5-9B thinking 43.2% $0.70
Gemini 2.5 Flash-Lite Sep 2025 max 42.5% $2.72
Nemotron 3.5 Super VL max 37.5% $1.26
Mistral Medium 3.5 high 36.5% $7.77
Qwen3-VL-235B-A22B-Instruct default 36% $0.39
Claude Haiku 4.5 max 31.2% $3.02
Qwen2.5-VL-72B default 26.3% $0.31
GPT-4o mini default 25.7% $0.56

14 models have held number one on the official leaderboard since 2023

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

OpenAI Qwen Meta Google DeepMind Anthropic
Own leaderboard Passed 85% in Apr 2026, by GPT-5.5

Source: Official Razavi-Bench leaderboard

GPT-5.5 on the official leaderboard passed 85% in April 2026

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

Plot with other benchmarks
On the benchmark’s own leaderboard Best so far
0% 20% 40% 60% 80% 100% 2025 2026 2027 Near the ceiling (85%) Gemini 4 Carbon Claude Opus 5 GPT-5.5 Gemini 3 Flash GPT-5.2 Gemini 2.5 Pro o3 0% 20% 40% 60% 80% 100% 2025 2026 2027 Near the ceiling (85%) Gemini 4 Carbon Claude Fable 5

Source: Official Razavi-Bench 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 Engineering and CAD

All 112 in Engineering and CAD
  • EEBench In use 13 electrical-engineering circuit and PCB design tasks graded by deterministic build checks, simulation and BOM constraints. No LLM judge. Gemini 4 Carbon 83.9%
  • 3DCodeBench In use Agentic procedural 3D modeling via code: author engine-ready 3D assets through software APIs. Graded for executability and shape fidelity. LongCat-2.5 55.4%
  • CADGenBench Only in system cards Agents construct CAD models from design prompts. The generation split is graded for executability and geometric correctness. Claude Fable 5 45.1%
  • PartBench Only in system cards Parametric mechanical part design and CAD-style generation, graded for geometric and manufacturing fidelity. Grok 4.6 58.5%
  • CircuitRubric Not in model cards Models write analog circuit netlists from design requests. A program compares transistor connections and relative device sizes with accepted reference circuits. Claude Opus 4.8 96%
  • AnalogCoder Not in model cards Models write Python programs that build analog circuits. A simulator checks whether each circuit performs its requested function. The authors report first-attempt success and how many tasks each model solves. Llama 3.1-405B 83.3%

Data audit

All 10 checks pass. Every benchmark

Catalog record 2 pass
  • Pass Task access All tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 4 pass 1 note
  • Pass Independent runs Scores from the official leaderboard (73 entries).
  • Pass Models scored 63 models we track, plus 10 other entries on its leaderboard.
  • Pass Newest score The newest is from November 2026, for Claude Fable 5.2.
  • Pass Frontier models 7 of the top 10 have a score, led by Claude Fable 5.2 (#2).
  • Note Headroom Saturated. The best is 98.3%, past 85% since April 2026.
Leaderboard import 3 pass
  • Pass Importer We read the board on razavi-bench.tokenzhang.com: 63 rows for models we track and 10 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.

How we check