Benchmark Almanac

Adversarial Humanities Benchmark

Claude Sonnet 4.6 9.20% leads since April 2026
Not in model cards

Models respond to harmful requests rewritten in literary and scholarly styles. A judge measures how often the rewrites elicit unsafe answers.

Released
April 2026
Kind
Jailbreaks and prompt injection
Level
Everyday
Scored
Graded by a model
Answers
Open-ended
Tasks
All public
Measures
Propensity
Behavior
Harmful use
Language
English
In and out
Text → Text
Interaction
Single turn
Refresh
One fixed set
Run by
Anyone
In views
Content safety
Strongest model tested
GPT-5.4 #28 on the capability index
Last model added
April 2026
Scores refreshed
November 20th, 2026
31
Models scored in its paper

Claude Sonnet 4.6 leads with 9.20% among models released by April 2026

Best score per model from the benchmark's paper, read on November 20th, 2026. A paper’s table doesn’t add newer models. Where a model ran with several settings or harnesses, we keep its best.

The newest model with a score here came out in April 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

  1. 1 Claude Sonnet 4.6 9.20%
  2. 2 Claude Opus 4.6 15.6%
  3. 3 GPT-5.4 Mini 22.5%
  4. 4 Claude Haiku 4.5 25.5%
  5. 5 MiniMax-M2.7 29.2%
  6. 6 Qwen3.5-27B 29.3%
  7. 7 GPT-5.4 30.2%
  8. 8 Qwen3.5-Flash 37%
  9. 9 Qwen3.5-Plus 38.6%
  10. 10 Qwen 3.6 Plus 41.1%
  11. 11 Grok 4.1 Fast 54.2%
  12. 12 gpt-oss-120b 55.6%
  13. 13 gpt-oss-20b 59.5%
  14. 14 Llama 4 Scout 61.5%
  15. 15 Seed2.0 Lite (0215) 62.5%
  16. 16 Llama 4 Maverick 63.6%
  17. 17 Qwen3-Max 63.8%
  18. 18 Mistral Small 4 65.2%
  19. 19 Doubao-Seed-1.6 66.4%
  20. 20 Gemma 4 31B IT 68.4%
  21. 21 Seed2.0 Mini 72%
  22. 22 GLM-5.1 72.3%
  23. 23 DeepSeek-R1 May 2025 72.6%
  24. 24 Gemini 3.1 Pro 73.2%
  25. 25 Kimi K2.5 74.1%
  26. 26 DeepSeek-V3.2 74.5%
  27. 27 GLM-5-Turbo 75.3%
  28. 28 Grok 4.20 77.2%
  29. 29 Gemini 3.1 Flash-Lite 77.5%
  30. 30 Mistral Large 3 79.8%
  31. 31 Gemini 3 Flash 81%

Source: The Adversarial Humanities Benchmark paper

#ModelScoreSettingDateSource
1 Claude Sonnet 4.6 9.20% Feb 2026 Official leaderboard
2 Claude Opus 4.6 15.6% Feb 2026 Official leaderboard
3 GPT-5.4 Mini 22.5% Mar 2026 Official leaderboard
4 Claude Haiku 4.5 25.5% Oct 2025 Official leaderboard
5 MiniMax-M2.7 29.2% Mar 2026 Official leaderboard
6 Qwen3.5-27B 29.3% Feb 2026 Official leaderboard
7 GPT-5.4 30.2% Mar 2026 Official leaderboard
8 Qwen3.5-Flash 37% Feb 2026 Official leaderboard
9 Qwen3.5-Plus 38.6% Feb 2026 Official leaderboard
10 Qwen 3.6 Plus 41.1% Mar 2026 Official leaderboard
11 Grok 4.1 Fast 54.2% Nov 2025 Official leaderboard
12 gpt-oss-120b 55.6% Aug 2025 Official leaderboard
13 gpt-oss-20b 59.5% Aug 2025 Official leaderboard
14 Llama 4 Scout 61.5% Apr 2025 Official leaderboard
15 Seed2.0 Lite (0215) 62.5% Feb 2026 Official leaderboard
16 Llama 4 Maverick 63.6% Apr 2025 Official leaderboard
17 Qwen3-Max 63.8% Sep 2025 Official leaderboard
18 Mistral Small 4 65.2% Mar 2026 Official leaderboard
19 Doubao-Seed-1.6 66.4% Jun 2025 Official leaderboard
20 Gemma 4 31B IT 68.4% Apr 2026 Official leaderboard
21 Seed2.0 Mini 72% Feb 2026 Official leaderboard
22 GLM-5.1 72.3% Apr 2026 Official leaderboard
23 DeepSeek-R1 May 2025 72.6% May 2025 Official leaderboard
24 Gemini 3.1 Pro 73.2% Feb 2026 Official leaderboard
25 Kimi K2.5 74.1% Jan 2026 Official leaderboard
26 DeepSeek-V3.2 74.5% Dec 2025 Official leaderboard
27 GLM-5-Turbo 75.3% Mar 2026 Official leaderboard
28 Grok 4.20 77.2% Feb 2026 Official leaderboard
29 Gemini 3.1 Flash-Lite 77.5% Mar 2026 Official leaderboard
30 Mistral Large 3 79.8% Dec 2025 Official leaderboard
31 Gemini 3 Flash 81% Dec 2025 Official leaderboard

Rankings by part

We read 15 parts from the benchmark’s paper, each ranking models on one part of the test, such as vcr, ncr and dfm. Claude Sonnet 4.6 leads 8 of them, and 3 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.

vcr

31 models we track have a score here.

  1. 1 Claude Opus 4.6 ↑1 2.20%
  2. 2 Claude Sonnet 4.6 ↓1 2.20%
  3. 3 Claude Haiku 4.5 ↑1 23.9%
  4. 4 GPT-5.4 Mini ↓1 23.9%
  5. 5 GPT-5.4 ↑2 26.7%
  6. 6 MiniMax-M2.7 ↓1 35%
  7. 7 Qwen3.5-Flash ↑1 37%
  8. 8 Qwen3.5-27B ↓2 37%
  9. 9 Qwen3.5-Plus 41.3%
  10. 10 Qwen 3.6 Plus 52.2%

Number one over time

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

Anthropic
Benchmark paper Not tested since Apr 2026

Source: The Adversarial Humanities Benchmark paper

Scores over time

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

Plot with other benchmarks
In the benchmark’s paper Best so far
0% 20% 40% 60% 80% 100% 2025 2026 Claude Sonnet 4.6 Claude Opus 4.6 Claude Haiku 4.5 gpt-oss-120b Llama 4 Scout 0% 20% 40% 60% 80% 100% 2025 2026 Claude Sonnet 4.6 Claude Opus 4.6

Source: Official Adversarial Humanities Benchmark 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 Safety and alignment

All 636 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%
  • Covert misbehavior (Thinking Machines) In use Agentic tasks that check whether a model misbehaves and conceals it. Scored by the rate of covert misbehavior. Inkling 2 Mini 8.30%
  • Destructive actions (Thinking Machines) In use Agentic tasks that check whether a model takes destructive actions it was not authorized to take. Scored by the rate of destructive actions. Inkling 2 Small 4.70%
  • Misbehavior (Thinking Machines) In use Agentic tasks that check whether a model fabricates results, ignores instructions or hides what it did when a task is hard. Scored by the rate of misbehavior. Inkling 2 Mini 9.80%

Data audit

8 of 9 checks pass, 1 warning. Look first at the Newest score check. The newest came 7 months ago, in April 2026, for GLM-5.1. Every benchmark

Catalog record 2 pass
  • Pass Task access All tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 1 warning 2 pass 2 notes
  • Pass Independent runs Scores from the benchmark’s paper (31 entries).
  • Pass Models scored 31 models we track.
  • Warning Newest score The newest came 7 months ago, in April 2026, for GLM-5.1.
  • Note Frontier models None of the top 10 has a score. The strongest tested is GPT-5.4 (#28).
  • Note Headroom Lower is better, so there’s no ceiling to reach.
Leaderboard import 3 pass
  • Pass Importer We read the board on arxiv.org: 31 rows for models we track 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