Benchmark Almanac

BioTIER Refuse

Claude Opus 5.5 99.1% leads since September 2026
Claude Opus 5.5 99.1% leads since September 2026
Only in system cards

Models receive 398 dangerous biology prompts. The score measures the percentage correctly refused.

One track of BioTIER. The BioTIER record keeps lab scores that combine tracks.

Made by
SecureBio
Released
September 2026
Kind
Refusing harmful requests
Level
Expert
Scored
Graded by a model
Answers
Open-ended
Tasks
Kept private
Strongest model tested
GPT-6 Astra #3 on the capability index
Last in a system card
October 2026
Last model added
October 2026
Scores refreshed
November 12th, 2026
77
Entries on its own leaderboard, 69 models we track
6
System cards report it, from 3 labs

Claude Opus 5.5 leads with 99.1%

Best score per entry from the benchmark's own leaderboard, read on November 12th, 2026. 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.

Claude Opus 5.5 leads with 99.1%, claims included

Best score per model from every source, including labs’ own numbers and claims Epoch AI collected from reports and papers. Labs pick their own settings, so a claim can beat an independent run of the same model.

77 entries. Scroll sideways for the rest.

  1. 99.1 Claude Opus 5.5
  2. 98.2 Claude Sonnet 4.5
  3. 97.9 Claude Haiku 5.5
  4. 96.1 Claude Sonnet 5
  5. 95.9 Claude Opus 4.8
  6. 95.7 Claude Sonnet 5.5
  7. 95.5 Muse Spark Open
  8. 95.2 Claude Sonnet 4.6
  9. 94.9 Claude Opus 4.7
  10. 94.9 Claude Opus 5
  11. 94.4 Claude Opus 4.6
  12. 92.6 Claude Opus 4.5
  13. 87.8 Gemini 4 Argon
  14. 81.7 Gemini 3.5 Flash
  15. 81.1 Gemini 3.6 Flash
  16. 80.3 Grok 4.7
  17. 79.8 GPT-6 Sol
  18. 79.5 GPT-6 Luna
  19. 79.1 GPT-6.1 Terra
  20. 78.2 Gemini 3.7 Flash
  21. 78.1 GPT-6 Astra
  22. 78.0 Gemini 3.8 Flash
  23. 77.8 GPT-6.1 Sol
  24. 76.9 GPT-5.4
  25. 76.3 Muse Glimmer 30B
  26. 76.2 Grok 4.5
  27. 76.1 Grok 4.6
  28. 75.4 GPT-5.2
  29. 75.3 GPT-5.5
  30. 74.0 GPT-5.6 Sol
  31. 73.6 GPT-5.5 Pro
  32. 72.1 GPT-5.6 Luna
  33. 71.7 GPT-5.6 Terra
  34. 68.7 Gemini 3.1 Pro (Feb 26' Preview)
  35. 66.9 Grok 4.20
  36. 66.0 Gemini 3 Flash (Dec '25 Preview)
  37. 61.6 Gemma 4 31B IT
  38. 61.4 GPT-5.5 Rosalind
  39. 61.3 Beam 1.0
  40. 45.4 gpt-oss-120b
  41. 42.0 Claude Haiku 4.5
  42. 39.3 gpt-oss-20b
  43. 35.7 Qwen3.5 397B-A17B
  44. 33.9 GLM-5
  45. 32.8 Qwen3.7-Max
  46. 32.6 GLM-5.2
  47. 32.6 Kimi K3.1 Pro
  48. 32.5 GLM-5.3
  49. 32.4 Nemotron 3 Ultra 550B A55B
  50. 32.2 Mistral Large 4
  51. 32.1 Nemotron 3.5 Super VL
  52. 31.6 Qwen3.8-2.4T-A95B
  53. 30.0 Inkling
  54. 27.3 Kimi K3
  55. 25.9 DeepSeek V4.1 Flash 1017
  56. 24.9 Inkling-Small
  57. 23.9 Nemotron 3 Super 120B A12B
  58. 23.7 GLM-5.1
  59. 22.1 GPT-4o
  60. 21.9 Kimi K2.6
  61. 21.9 Qwen 3.8 27B
  62. 21.2 Llama 3.3 70B
  63. 21.2 Kimi K2.5
  64. 21.1 Nemotron 3.5 Lightning 30B-A3B
  65. 20.9 Qwen3-235B-A22B-Thinking
  66. 20.2 Nemotron 3 Nano 30B-A3B
  67. 17.6 DeepSeek V4 Pro 0813
  68. 16.7 GPT-4.1
  69. 16.3 Kimi K2 Thinking
  70. 15.6 DeepSeek-R1
  71. 15.4 GPT-4o mini (snapshot unspecified)
  72. 14.5 GPT-4o
  73. 10.4 GPT-4.1 mini
  74. 10.3 Mistral Medium 3.1
  75. 10.2 DeepSeek-V3.2
  76. 9.30 Gemini 2.5 Flash (snapshot unspecified)
  77. 9.20 Mistral Large 3
  1. 1 Claude Opus 5.5 max 99.1%
  2. 2 Claude Sonnet 4.5 16k 98.2%
  3. 3 Claude Haiku 5.5 max 97.9%
  4. 4 Claude Sonnet 5 max 96.1%
  5. 5 Claude Opus 4.8 max 95.9%
  6. 6 Claude Sonnet 5.5 max 95.7%
  7. 7 Muse Spark Open xhigh 95.5%
  8. 8 Claude Sonnet 4.6 max 95.2%
  9. 9 Claude Opus 4.7 max 94.9%
  10. 10 Claude Opus 5 max 94.9%
  11. 11 Claude Opus 4.6 max 94.4%
  12. 12 Claude Opus 4.5 16k 92.6%
  13. 13 Gemini 4 Argon high 87.8%
  14. 14 Gemini 3.5 Flash high 81.7%
  15. 15 Gemini 3.6 Flash high 81.1%
  16. 16 Grok 4.7 xhigh 80.3%
  17. 17 GPT-6 Sol max 79.8%
  18. 18 GPT-6 Luna max 79.5%
  19. 19 GPT-6.1 Terra max 79.1%
  20. 20 Gemini 3.7 Flash high 78.2%
  21. 21 GPT-6 Astra max 78.1%
  22. 22 Gemini 3.8 Flash high 78%
  23. 23 GPT-6.1 Sol max 77.8%
  24. 24 GPT-5.4 xhigh 76.9%
  25. 25 Muse Glimmer 30B xhigh 76.3%
  26. 26 Grok 4.5 high 76.2%
  27. 27 Grok 4.6 xhigh 76.1%
  28. 28 GPT-5.2 xhigh 75.4%
  29. 29 GPT-5.5 xhigh 75.3%
  30. 30 GPT-5.6 Sol max 74%
  31. 31 GPT-5.5 Pro xhigh 73.6%
  32. 32 GPT-5.6 Luna max 72.1%
  33. 33 GPT-5.6 Terra max 71.7%
  34. 34 Gemini 3.1 Pro (Feb 26' Preview) high 68.7%
  35. 35 Grok 4.20 reasoning 66.9%
  36. 36 Gemini 3 Flash (Dec '25 Preview) high 66%
  37. 37 Gemma 4 31B IT 61.6%
  38. 38 GPT-5.5 Rosalind xhigh 61.4%
  39. 39 Beam 1.0 max 61.3%
  40. 40 gpt-oss-120b 45.4%
  41. 41 Claude Haiku 4.5 16k 42%
  42. 42 gpt-oss-20b 39.3%
  43. 43 Qwen3.5 397B-A17B 35.7%
  44. 44 GLM-5 33.9%
  45. 45 Qwen3.7-Max 32.8%
  46. 46 GLM-5.2 max 32.6%
  47. 47 Kimi K3.1 Pro max 32.6%
  48. 48 GLM-5.3 max 32.5%
  49. 49 Nemotron 3 Ultra 550B A55B 32.4%
  50. 50 Mistral Large 4 high 32.2%
  51. 51 Nemotron 3.5 Super VL 32.1%
  52. 52 Qwen3.8-2.4T-A95B xhigh 31.6%
  53. 53 Inkling xhigh 30%
  54. 54 Kimi K3 max 27.3%
  55. 55 DeepSeek V4.1 Flash 1017 max 25.9%
  56. 56 Inkling-Small xhigh 24.9%
  57. 57 Nemotron 3 Super 120B A12B 23.9%
  58. 58 GLM-5.1 23.7%
  59. 59 GPT-4o Nov 2024 22.1%
  60. 60 Kimi K2.6 21.9%
  61. 61 Qwen 3.8 27B xhigh 21.9%
  62. 62 Llama 3.3 70B 21.2%
  63. 63 Kimi K2.5 21.2%
  64. 64 Nemotron 3.5 Lightning 30B-A3B 21.1%
  65. 65 Qwen3-235B-A22B-Thinking Jul 2025 20.9%
  66. 66 Nemotron 3 Nano 30B-A3B 20.2%
  67. 67 DeepSeek V4 Pro 0813 max 17.6%
  68. 68 GPT-4.1 16.7%
  69. 69 Kimi K2 Thinking 16.3%
  70. 70 DeepSeek-R1 May 2025 15.6%
  71. 71 GPT-4o mini (snapshot unspecified) 15.4%
  72. 72 GPT-4o Aug 2024 14.5%
  73. 73 GPT-4.1 mini 10.4%
  74. 74 Mistral Medium 3.1 10.3%
  75. 75 DeepSeek-V3.2 10.2%
  76. 76 Gemini 2.5 Flash (snapshot unspecified) 16k 9.30%
  77. 77 Mistral Large 3 9.20%

Source: Official BioTIER Refuse leaderboard

73 models. Scroll sideways for the rest.

  1. 99.1 Claude Opus 5.5
  2. 98.7 Muse Wonder Full
  3. 98.2 Claude Sonnet 4.5
  4. 98.0 Muse Spark
  5. 97.9 Claude Haiku 5.5
  6. 97.7 Muse Spark 1.1
  7. 97.4 Muse Spark 1.3
  8. 96.1 Claude Sonnet 5
  9. 95.9 Claude Opus 4.8
  10. 95.7 Claude Sonnet 5.5
  11. 95.5 Muse Spark Open
  12. 95.2 Claude Sonnet 4.6
  13. 94.9 Claude Opus 4.7
  14. 94.9 Claude Opus 5
  15. 94.4 Claude Opus 4.6
  16. 92.6 Claude Opus 4.5
  17. 87.8 Gemini 4 Argon
  18. 82.9 GPT-6 Astra
  19. 81.7 Gemini 3.5 Flash
  20. 81.1 Gemini 3.6 Flash
  21. 80.3 Grok 4.7
  22. 79.8 GPT-6 Sol
  23. 79.5 GPT-6 Luna
  24. 79.1 GPT-6.1 Terra
  25. 78.2 Gemini 3.7 Flash
  26. 78.0 Gemini 3.8 Flash
  27. 77.8 GPT-6.1 Sol
  28. 76.9 GPT-5.4
  29. 76.2 Grok 4.5
  30. 76.1 Grok 4.6
  31. 75.4 GPT-5.2
  32. 75.3 GPT-5.5
  33. 74.0 GPT-5.6 Sol
  34. 73.6 GPT-5.5 Pro
  35. 72.1 GPT-5.6 Luna
  36. 71.7 GPT-5.6 Terra
  37. 66.9 Grok 4.20
  38. 64.6 Beam 1.0
  39. 61.6 Gemma 4 31B IT
  40. 45.4 gpt-oss-120b
  41. 42.0 Claude Haiku 4.5
  42. 39.3 gpt-oss-20b
  43. 35.7 Qwen3.5 397B-A17B
  44. 33.9 GLM-5
  45. 32.8 Qwen3.7-Max
  46. 32.6 GLM-5.2
  47. 32.6 Kimi K3.1 Pro
  48. 32.5 GLM-5.3
  49. 32.2 Mistral Large 4
  50. 32.1 Nemotron 3.5 Super VL
  51. 31.6 Qwen3.8-2.4T-A95B
  52. 30.0 Inkling
  53. 27.3 Kimi K3
  54. 25.9 DeepSeek V4.1 Flash 1017
  55. 24.9 Inkling-Small
  56. 23.7 GLM-5.1
  57. 22.1 GPT-4o
  58. 21.9 Kimi K2.6
  59. 21.9 Qwen 3.8 27B
  60. 21.2 Llama 3.3 70B
  61. 21.2 Kimi K2.5
  62. 21.1 Nemotron 3.5 Lightning 30B-A3B
  63. 20.9 Qwen3-235B-A22B-Thinking
  64. 20.2 Nemotron 3 Nano 30B-A3B
  65. 17.6 DeepSeek V4 Pro 0813
  66. 16.7 GPT-4.1
  67. 16.3 Kimi K2 Thinking
  68. 15.6 DeepSeek-R1
  69. 14.5 GPT-4o
  70. 10.4 GPT-4.1 mini
  71. 10.3 Mistral Medium 3.1
  72. 10.2 DeepSeek-V3.2
  73. 9.20 Mistral Large 3
  1. 1 Claude Opus 5.5 max 99.1%
  2. 2 Muse Wonder Full 98.7%
  3. 3 Claude Sonnet 4.5 16k 98.2%
  4. 4 Muse Spark 98%
  5. 5 Claude Haiku 5.5 max 97.9%
  6. 6 Muse Spark 1.1 97.7%
  7. 7 Muse Spark 1.3 97.4%
  8. 8 Claude Sonnet 5 max 96.1%
  9. 9 Claude Opus 4.8 max 95.9%
  10. 10 Claude Sonnet 5.5 max 95.7%
  11. 11 Muse Spark Open xhigh 95.5%
  12. 12 Claude Sonnet 4.6 max 95.2%
  13. 13 Claude Opus 4.7 max 94.9%
  14. 14 Claude Opus 5 max 94.9%
  15. 15 Claude Opus 4.6 max 94.4%
  16. 16 Claude Opus 4.5 16k 92.6%
  17. 17 Gemini 4 Argon high 87.8%
  18. 18 GPT-6 Astra 82.9%
  19. 19 Gemini 3.5 Flash high 81.7%
  20. 20 Gemini 3.6 Flash high 81.1%
  21. 21 Grok 4.7 xhigh 80.3%
  22. 22 GPT-6 Sol max 79.8%
  23. 23 GPT-6 Luna max 79.5%
  24. 24 GPT-6.1 Terra max 79.1%
  25. 25 Gemini 3.7 Flash high 78.2%
  26. 26 Gemini 3.8 Flash high 78%
  27. 27 GPT-6.1 Sol max 77.8%
  28. 28 GPT-5.4 xhigh 76.9%
  29. 29 Grok 4.5 high 76.2%
  30. 30 Grok 4.6 xhigh 76.1%
  31. 31 GPT-5.2 xhigh 75.4%
  32. 32 GPT-5.5 xhigh 75.3%
  33. 33 GPT-5.6 Sol max 74%
  34. 34 GPT-5.5 Pro xhigh 73.6%
  35. 35 GPT-5.6 Luna max 72.1%
  36. 36 GPT-5.6 Terra max 71.7%
  37. 37 Grok 4.20 reasoning 66.9%
  38. 38 Beam 1.0 64.6%
  39. 39 Gemma 4 31B IT 61.6%
  40. 40 gpt-oss-120b 45.4%
  41. 41 Claude Haiku 4.5 16k 42%
  42. 42 gpt-oss-20b 39.3%
  43. 43 Qwen3.5 397B-A17B 35.7%
  44. 44 GLM-5 33.9%
  45. 45 Qwen3.7-Max 32.8%
  46. 46 GLM-5.2 max 32.6%
  47. 47 Kimi K3.1 Pro max 32.6%
  48. 48 GLM-5.3 max 32.5%
  49. 49 Mistral Large 4 high 32.2%
  50. 50 Nemotron 3.5 Super VL 32.1%
  51. 51 Qwen3.8-2.4T-A95B xhigh 31.6%
  52. 52 Inkling xhigh 30%
  53. 53 Kimi K3 max 27.3%
  54. 54 DeepSeek V4.1 Flash 1017 max 25.9%
  55. 55 Inkling-Small xhigh 24.9%
  56. 56 GLM-5.1 23.7%
  57. 57 GPT-4o Nov 2024 22.1%
  58. 58 Kimi K2.6 21.9%
  59. 59 Qwen 3.8 27B xhigh 21.9%
  60. 60 Llama 3.3 70B 21.2%
  61. 61 Kimi K2.5 21.2%
  62. 62 Nemotron 3.5 Lightning 30B-A3B 21.1%
  63. 63 Qwen3-235B-A22B-Thinking Jul 2025 20.9%
  64. 64 Nemotron 3 Nano 30B-A3B 20.2%
  65. 65 DeepSeek V4 Pro 0813 max 17.6%
  66. 66 GPT-4.1 16.7%
  67. 67 Kimi K2 Thinking 16.3%
  68. 68 DeepSeek-R1 May 2025 15.6%
  69. 69 GPT-4o Aug 2024 14.5%
  70. 70 GPT-4.1 mini 10.4%
  71. 71 Mistral Medium 3.1 10.3%
  72. 72 DeepSeek-V3.2 10.2%
  73. 73 Mistral Large 3 9.20%

Sources: Official BioTIER Refuse leaderboard and labs’ model cards

#EntryScoreSettingDateSource
1 Claude Opus 5.5 99.1% max Sep 2026 Official leaderboard
2 Claude Sonnet 4.5 98.2% 16k Sep 2025 Official leaderboard
3 Claude Haiku 5.5 97.9% max Oct 2026 Official leaderboard
4 Claude Sonnet 5 96.1% max Jun 2026 Official leaderboard
5 Claude Opus 4.8 95.9% max May 2026 Official leaderboard
6 Claude Sonnet 5.5 95.7% max Sep 2026 Official leaderboard
7 Muse Spark Open 95.5% xhigh Oct 2026 Official leaderboard
8 Claude Sonnet 4.6 95.2% max Feb 2026 Official leaderboard
9 Claude Opus 4.7 94.9% max Apr 2026 Official leaderboard
10 Claude Opus 5 94.9% max Jul 2026 Official leaderboard
11 Claude Opus 4.6 94.4% max Feb 2026 Official leaderboard
12 Claude Opus 4.5 92.6% 16k Nov 2025 Official leaderboard
13 Gemini 4 Argon 87.8% high Sep 2026 Official leaderboard
14 Gemini 3.5 Flash 81.7% high May 2026 Official leaderboard
15 Gemini 3.6 Flash 81.1% high Jul 2026 Official leaderboard
16 Grok 4.7 80.3% xhigh Sep 2026 Official leaderboard
17 GPT-6 Sol 79.8% max Sep 2026 Official leaderboard
18 GPT-6 Luna 79.5% max Sep 2026 Official leaderboard
19 GPT-6.1 Terra 79.1% max Oct 2026 Official leaderboard
20 Gemini 3.7 Flash 78.2% high Aug 2026 Official leaderboard
21 GPT-6 Astra 78.1% max Sep 2026 Official leaderboard
22 Gemini 3.8 Flash 78% high Sep 2026 Official leaderboard
23 GPT-6.1 Sol 77.8% max Sep 2026 Official leaderboard
24 GPT-5.4 76.9% xhigh Mar 2026 Official leaderboard
25 Muse Glimmer 30B 76.3% xhigh Official leaderboard
26 Grok 4.5 76.2% high Jul 2026 Official leaderboard
27 Grok 4.6 76.1% xhigh Aug 2026 Official leaderboard
28 GPT-5.2 75.4% xhigh Dec 2025 Official leaderboard
29 GPT-5.5 75.3% xhigh Apr 2026 Official leaderboard
30 GPT-5.6 Sol 74% max Jul 2026 Official leaderboard
31 GPT-5.5 Pro 73.6% xhigh Apr 2026 Official leaderboard
32 GPT-5.6 Luna 72.1% max Jul 2026 Official leaderboard
33 GPT-5.6 Terra 71.7% max Jul 2026 Official leaderboard
34 Gemini 3.1 Pro (Feb 26' Preview) 68.7% high Official leaderboard
35 Grok 4.20 66.9% reasoning Feb 2026 Official leaderboard
36 Gemini 3 Flash (Dec '25 Preview) 66% high Official leaderboard
37 Gemma 4 31B IT 61.6% Apr 2026 Official leaderboard
38 GPT-5.5 Rosalind 61.4% xhigh Official leaderboard
39 Beam 1.0 61.3% max Oct 2026 Official leaderboard
40 gpt-oss-120b 45.4% Aug 2025 Official leaderboard
41 Claude Haiku 4.5 42% 16k Oct 2025 Official leaderboard
42 gpt-oss-20b 39.3% Aug 2025 Official leaderboard
43 Qwen3.5 397B-A17B 35.7% Feb 2026 Official leaderboard
44 GLM-5 33.9% Feb 2026 Official leaderboard
45 Qwen3.7-Max 32.8% May 2026 Official leaderboard
46 GLM-5.2 32.6% max Jun 2026 Official leaderboard
47 Kimi K3.1 Pro 32.6% max Oct 2026 Official leaderboard
48 GLM-5.3 32.5% max Aug 2026 Official leaderboard
49 Nemotron 3 Ultra 550B A55B 32.4% Official leaderboard
50 Mistral Large 4 32.2% high Oct 2026 Official leaderboard
51 Nemotron 3.5 Super VL 32.1% Oct 2026 Official leaderboard
52 Qwen3.8-2.4T-A95B 31.6% xhigh Aug 2026 Official leaderboard
53 Inkling 30% xhigh Jul 2026 Official leaderboard
54 Kimi K3 27.3% max Jul 2026 Official leaderboard
55 DeepSeek V4.1 Flash 1017 25.9% max Oct 2026 Official leaderboard
56 Inkling-Small 24.9% xhigh Jul 2026 Official leaderboard
57 Nemotron 3 Super 120B A12B 23.9% Official leaderboard
58 GLM-5.1 23.7% Apr 2026 Official leaderboard
59 GPT-4o Nov 2024 22.1% Nov 2024 Official leaderboard
60 Kimi K2.6 21.9% Apr 2026 Official leaderboard
61 Qwen 3.8 27B 21.9% xhigh Aug 2026 Official leaderboard
62 Llama 3.3 70B 21.2% Dec 2024 Official leaderboard
63 Kimi K2.5 21.2% Jan 2026 Official leaderboard
64 Nemotron 3.5 Lightning 30B-A3B 21.1% Aug 2026 Official leaderboard
65 Qwen3-235B-A22B-Thinking Jul 2025 20.9% Jul 2025 Official leaderboard
66 Nemotron 3 Nano 30B-A3B 20.2% Dec 2025 Official leaderboard
67 DeepSeek V4 Pro 0813 17.6% max Aug 2026 Official leaderboard
68 GPT-4.1 16.7% Apr 2025 Official leaderboard
69 Kimi K2 Thinking 16.3% Nov 2025 Official leaderboard
70 DeepSeek-R1 May 2025 15.6% May 2025 Official leaderboard
71 GPT-4o mini (snapshot unspecified) 15.4% Official leaderboard
72 GPT-4o Aug 2024 14.5% Aug 2024 Official leaderboard
73 GPT-4.1 mini 10.4% Apr 2025 Official leaderboard
74 Mistral Medium 3.1 10.3% Aug 2025 Official leaderboard
75 DeepSeek-V3.2 10.2% Dec 2025 Official leaderboard
76 Gemini 2.5 Flash (snapshot unspecified) 9.30% 16k Official leaderboard
77 Mistral Large 3 9.20% Dec 2025 Official leaderboard
#ModelScoreSettingDateSource
1 Claude Opus 5.5 99.1% max Sep 2026 Official leaderboard
2 Muse Wonder Full 98.7% Oct 2026 Lab model card
3 Claude Sonnet 4.5 98.2% 16k Sep 2025 Official leaderboard
4 Muse Spark 98% Jun 2026 Lab model card
5 Claude Haiku 5.5 97.9% max Oct 2026 Official leaderboard
6 Muse Spark 1.1 97.7% Jul 2026 Lab model card
7 Muse Spark 1.3 97.4% Oct 2026 Lab model card
8 Claude Sonnet 5 96.1% max Jun 2026 Official leaderboard
9 Claude Opus 4.8 95.9% max May 2026 Official leaderboard
10 Claude Sonnet 5.5 95.7% max Sep 2026 Official leaderboard
11 Muse Spark Open 95.5% xhigh Oct 2026 Official leaderboard
12 Claude Sonnet 4.6 95.2% max Feb 2026 Official leaderboard
13 Claude Opus 4.7 94.9% max Apr 2026 Official leaderboard
14 Claude Opus 5 94.9% max Jul 2026 Official leaderboard
15 Claude Opus 4.6 94.4% max Feb 2026 Official leaderboard
16 Claude Opus 4.5 92.6% 16k Nov 2025 Official leaderboard
17 Gemini 4 Argon 87.8% high Sep 2026 Official leaderboard
18 GPT-6 Astra 82.9% Sep 2026 Lab model card
19 Gemini 3.5 Flash 81.7% high May 2026 Official leaderboard
20 Gemini 3.6 Flash 81.1% high Jul 2026 Official leaderboard
21 Grok 4.7 80.3% xhigh Sep 2026 Official leaderboard
22 GPT-6 Sol 79.8% max Sep 2026 Official leaderboard
23 GPT-6 Luna 79.5% max Sep 2026 Official leaderboard
24 GPT-6.1 Terra 79.1% max Oct 2026 Official leaderboard
25 Gemini 3.7 Flash 78.2% high Aug 2026 Official leaderboard
26 Gemini 3.8 Flash 78% high Sep 2026 Official leaderboard
27 GPT-6.1 Sol 77.8% max Sep 2026 Official leaderboard
28 GPT-5.4 76.9% xhigh Mar 2026 Official leaderboard
29 Grok 4.5 76.2% high Jul 2026 Official leaderboard
30 Grok 4.6 76.1% xhigh Aug 2026 Official leaderboard
31 GPT-5.2 75.4% xhigh Dec 2025 Official leaderboard
32 GPT-5.5 75.3% xhigh Apr 2026 Official leaderboard
33 GPT-5.6 Sol 74% max Jul 2026 Official leaderboard
34 GPT-5.5 Pro 73.6% xhigh Apr 2026 Official leaderboard
35 GPT-5.6 Luna 72.1% max Jul 2026 Official leaderboard
36 GPT-5.6 Terra 71.7% max Jul 2026 Official leaderboard
37 Grok 4.20 66.9% reasoning Feb 2026 Official leaderboard
38 Beam 1.0 64.6% Oct 2026 Lab model card
39 Gemma 4 31B IT 61.6% Apr 2026 Official leaderboard
40 gpt-oss-120b 45.4% Aug 2025 Official leaderboard
41 Claude Haiku 4.5 42% 16k Oct 2025 Official leaderboard
42 gpt-oss-20b 39.3% Aug 2025 Official leaderboard
43 Qwen3.5 397B-A17B 35.7% Feb 2026 Official leaderboard
44 GLM-5 33.9% Feb 2026 Official leaderboard
45 Qwen3.7-Max 32.8% May 2026 Official leaderboard
46 GLM-5.2 32.6% max Jun 2026 Official leaderboard
47 Kimi K3.1 Pro 32.6% max Oct 2026 Official leaderboard
48 GLM-5.3 32.5% max Aug 2026 Official leaderboard
49 Mistral Large 4 32.2% high Oct 2026 Official leaderboard
50 Nemotron 3.5 Super VL 32.1% Oct 2026 Official leaderboard
51 Qwen3.8-2.4T-A95B 31.6% xhigh Aug 2026 Official leaderboard
52 Inkling 30% xhigh Jul 2026 Official leaderboard
53 Kimi K3 27.3% max Jul 2026 Official leaderboard
54 DeepSeek V4.1 Flash 1017 25.9% max Oct 2026 Official leaderboard
55 Inkling-Small 24.9% xhigh Jul 2026 Official leaderboard
56 GLM-5.1 23.7% Apr 2026 Official leaderboard
57 GPT-4o Nov 2024 22.1% Nov 2024 Official leaderboard
58 Kimi K2.6 21.9% Apr 2026 Official leaderboard
59 Qwen 3.8 27B 21.9% xhigh Aug 2026 Official leaderboard
60 Llama 3.3 70B 21.2% Dec 2024 Official leaderboard
61 Kimi K2.5 21.2% Jan 2026 Official leaderboard
62 Nemotron 3.5 Lightning 30B-A3B 21.1% Aug 2026 Official leaderboard
63 Qwen3-235B-A22B-Thinking Jul 2025 20.9% Jul 2025 Official leaderboard
64 Nemotron 3 Nano 30B-A3B 20.2% Dec 2025 Official leaderboard
65 DeepSeek V4 Pro 0813 17.6% max Aug 2026 Official leaderboard
66 GPT-4.1 16.7% Apr 2025 Official leaderboard
67 Kimi K2 Thinking 16.3% Nov 2025 Official leaderboard
68 DeepSeek-R1 May 2025 15.6% May 2025 Official leaderboard
69 GPT-4o Aug 2024 14.5% Aug 2024 Official leaderboard
70 GPT-4.1 mini 10.4% Apr 2025 Official leaderboard
71 Mistral Medium 3.1 10.3% Aug 2025 Official leaderboard
72 DeepSeek-V3.2 10.2% Dec 2025 Official leaderboard
73 Mistral Large 3 9.20% Dec 2025 Official leaderboard

Labs report scores for 6 models in their cards. Switch the scores above to rank them together.

2 models have held number one on the official leaderboard since its release in September 2026 2 models have held number one with claims included since its release in September 2026

Lab claims count from the day the card came out, independent runs from the model’s release. We draw one row each for the benchmark’s own leaderboard and what labs claim in their cards.

Anthropic Meta
Own leaderboard Passed 85% in Sep 2026, by Claude Sonnet 4.5
Claude Opus 5.5 99.1%
Lab claims Passed 85% in Sep 2026, by Muse Spark
Muse Wonder Full 98.7%

Sources: Official BioTIER Refuse leaderboard and labs’ model cards

Claude Sonnet 4.5 on the official leaderboard passed 85% in September 2025

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% 2025 2026 2027 Near the ceiling (85%) Claude Opus 5.5 Claude Sonnet 4.5 gpt-oss-120b GPT-4o GPT-4o 0% 20% 40% 60% 80% 100% 2025 2026 2027 Near the ceiling (85%) Claude Opus 5.5 gpt-oss-120b

Sources: Official BioTIER Refuse leaderboard and labs’ model cards

Labs report it only in system cards

We read the headline results of 522 launch posts from 52 labs, and none of them list this benchmark. Some system cards do, below.

6 system cards report it, from 3 labs

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
Oct 23, 2026 Muse Spark 1.3 system card 97.4 BioTIER v1 (English) refusal rate, Muse Spark 1.3 (API) with system mitigations (Table 2, Figure 5). Multilingual 96.9, adversarial 95.3 Bio and chem risk
Oct 12, 2026 Muse Wonder Full model card 98.7 BioTIER English refusal rate (CBRN Safety Evaluations table, section 4). Multilingual 98.7, adversarial 98.2 Bio and chem risk
Oct 5, 2026 Beam 1.0 technical report 64.6 BioTIER refuse split accuracy Bio and chem risk
Sep 3, 2026 GPT-6 Astra system card 82.9 SecureBio BioTIER, safeguarded configuration refused requests designated for refusal (benign allowed 95.3%) (10.2.4.2) Safety
Jul 9, 2026 Muse Spark 1.1 system card 97.7 Refusals: BioTIER v1 refusal rate (Table 2, Figure 6) Bio and chem risk
Jun 12, 2026 Muse Spark system card 98 Refusals: BioTIER refusal rate, deployed system (Table 2, Figure 9) Bio and chem risk

The lab's number comes out ahead in 2 of 2 pairs that differ

Each score a lab reported for its own model, beside the same model's best score on the leaderboard. Gaps mostly come from different harnesses, effort levels and task subsets. Every benchmark.

Model Lab says Board says Gap
GPT-6 Astra 82.9% SecureBio BioTIER, safeguarded configuration refused requests designated for refusal (benign allowed 95.3%) (10.2.4.2) 78.1% Leaderboard, max +4.8 pts
Beam 1.0 64.6% BioTIER refuse split accuracy 61.3% Leaderboard, max +3.3 pts

More in Safety and alignment

All 465 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%
  • 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. Gemini 4 Carbon 0%
  • DistressBench In use Models reply as a chatbot to 718 simulated users in psychological crisis, in single-turn and multi-turn chats. Three judge models score replies against clinician-written rubrics. Muse Spark 1.2 88.4%
  • Lakera B3 (Backbone Breaker Benchmark) In use Crowdsourced prompt-injection attacks hit backbone LLMs inside agent threat snapshots at three defense levels. The score is attack resistance, 100 minus the board's vulnerability percentage. Claude Opus 4.5 89%
  • IHEval In use Models face conflicts between system, user and tool instructions. Reports a composite score for following the higher-priority instruction. GPT-4o 70%