Benchmark Almanac

HumanAgencyBench

Claude Sonnet 4.6 71.3% leads since February 2026
Not in model cards

Models respond to 3,000 prompts about six ways assistants can support human agency, including asking clarifying questions and leaving important decisions to users.

In September 2026, the October 2024 Claude 3.5 Sonnet checkpoint led at 54.6% across six behaviors. The comparison includes GPT, Gemini, Grok, and Llama models.

Released
September 2025
Kind
Deception and alignment
Level
Everyday
Scored
Graded by a model
Answers
Open-ended
Tasks
All public
Measures
Propensity
Language
English
In and out
Text → Text
Interaction
Single turn
Refresh
One fixed set
Run by
Anyone
Strongest model tested
GPT-5.5 #21 on the capability index
Last model added
April 2026
Scores refreshed
November 20th, 2026
37
Entries on its own leaderboard, 34 models we track

Claude Sonnet 4.6 leads with 71.3% among models released by April 2026 Claude Sonnet 4.6 leads with 71.3% among models released by April 2026

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. 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.

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 71.3%
  2. 2 Claude Opus 4.5 68.5%
  3. 3 Claude Opus 4.7 65.1%
  4. 4 Claude Opus 4.6 63.2%
  5. 5 Claude Haiku 4.5 60.9%
  6. 6 Claude 3.5 Sonnet (October 2024) 54.6%
  7. 7 Claude Opus 4.1 53.4%
  8. 8 GPT-5 High reasoning effort 52.7%
  9. 9 Claude Sonnet 4 45.9%
  10. 10 Claude 3.5 Sonnet 44.1%
  11. 11 GPT-5.4 43.4%
  12. 12 Grok 4 43.4%
  13. 13 Claude 3.7 Sonnet 42.2%
  14. 14 Claude 3.5 Haiku 42.2%
  15. 15 Kimi K2.6 41.4%
  16. 16 Gemini 3.1 Pro 39%
  17. 17 GPT-5.5 38.9%
  18. 18 Gemini 2.5 Flash Apr 2025 38.2%
  19. 19 Gemini 2.5 Pro Mar 2025 35.9%
  20. 20 Llama 4 Maverick 35.9%
  21. 21 o3 34.7%
  22. 22 Grok 3 32.8%
  23. 23 GPT-5.4 Mini 32.5%
  24. 24 GLM-5.1 31.8%
  25. 25 Claude 3 Opus 31.7%
  26. 26 DeepSeek-V4-Pro 30.6%
  27. 27 GPT-4.1 30.6%
  28. 28 Gemini 2.0 Flash Feb 2025 30.3%
  29. 29 DeepSeek-V3.2 29.4%
  30. 30 Llama 4 Scout 28.7%
  31. 31 GPT-4.1 mini 24.4%
  32. 32 gemini-1.5-flash (snapshot unspecified) 23.7%
  33. 33 o4-mini 23.6%
  34. 34 gpt-4o (snapshot unspecified) 23.3%
  35. 35 Claude 3 Haiku 21.8%
  36. 36 meta-llama-3-70b-instruct 21.1%
  37. 37 o3-mini 18.2%

Source: Official HumanAgencyBench leaderboard

Source: Official HumanAgencyBench leaderboard

#EntryScoreSettingDateSource
1 Claude Sonnet 4.6 71.3% Feb 2026 Official leaderboard
2 Claude Opus 4.5 68.5% Nov 2025 Official leaderboard
3 Claude Opus 4.7 65.1% Apr 2026 Official leaderboard
4 Claude Opus 4.6 63.2% Feb 2026 Official leaderboard
5 Claude Haiku 4.5 60.9% Oct 2025 Official leaderboard
6 Claude 3.5 Sonnet (October 2024) 54.6% Oct 2024 Official leaderboard
7 Claude Opus 4.1 53.4% Aug 2025 Official leaderboard
8 GPT-5 52.7% High reasoning effort Aug 2025 Official leaderboard
9 Claude Sonnet 4 45.9% May 2025 Official leaderboard
10 Claude 3.5 Sonnet 44.1% Jun 2024 Official leaderboard
11 GPT-5.4 43.4% Mar 2026 Official leaderboard
12 Grok 4 43.4% Jul 2025 Official leaderboard
13 Claude 3.7 Sonnet 42.2% Feb 2025 Official leaderboard
14 Claude 3.5 Haiku 42.2% Oct 2024 Official leaderboard
15 Kimi K2.6 41.4% Apr 2026 Official leaderboard
16 Gemini 3.1 Pro 39% Feb 2026 Official leaderboard
17 GPT-5.5 38.9% Apr 2026 Official leaderboard
18 Gemini 2.5 Flash Apr 2025 38.2% Apr 2025 Official leaderboard
19 Gemini 2.5 Pro Mar 2025 35.9% Mar 2025 Official leaderboard
20 Llama 4 Maverick 35.9% Apr 2025 Official leaderboard
21 o3 34.7% Apr 2025 Official leaderboard
22 Grok 3 32.8% Apr 2025 Official leaderboard
23 GPT-5.4 Mini 32.5% Mar 2026 Official leaderboard
24 GLM-5.1 31.8% Apr 2026 Official leaderboard
25 Claude 3 Opus 31.7% Feb 2024 Official leaderboard
26 DeepSeek-V4-Pro 30.6% Apr 2026 Official leaderboard
27 GPT-4.1 30.6% Apr 2025 Official leaderboard
28 Gemini 2.0 Flash Feb 2025 30.3% Feb 2025 Official leaderboard
29 DeepSeek-V3.2 29.4% Dec 2025 Official leaderboard
30 Llama 4 Scout 28.7% Apr 2025 Official leaderboard
31 GPT-4.1 mini 24.4% Apr 2025 Official leaderboard
32 gemini-1.5-flash (snapshot unspecified) 23.7% Official leaderboard
33 o4-mini 23.6% Apr 2025 Official leaderboard
34 gpt-4o (snapshot unspecified) 23.3% Official leaderboard
35 Claude 3 Haiku 21.8% Mar 2024 Official leaderboard
36 meta-llama-3-70b-instruct 21.1% Official leaderboard
37 o3-mini 18.2% Jan 2025 Official leaderboard
#ModelScoreSettingDateSource
1 Claude Sonnet 4.6 71.3% Feb 2026 Official leaderboard
2 Claude Opus 4.5 68.5% Nov 2025 Official leaderboard
3 Claude Opus 4.7 65.1% Apr 2026 Official leaderboard
4 Claude Opus 4.6 63.2% Feb 2026 Official leaderboard
5 Claude Haiku 4.5 60.9% Oct 2025 Official leaderboard
6 Claude 3.5 Sonnet (October 2024) 54.6% Oct 2024 Official leaderboard
7 Claude Opus 4.1 53.4% Aug 2025 Official leaderboard
8 GPT-5 52.7% High reasoning effort Aug 2025 Official leaderboard
9 Claude Sonnet 4 45.9% May 2025 Official leaderboard
10 Claude 3.5 Sonnet 44.1% Jun 2024 Official leaderboard
11 GPT-5.4 43.4% Mar 2026 Official leaderboard
12 Grok 4 43.4% Jul 2025 Official leaderboard
13 Claude 3.7 Sonnet 42.2% Feb 2025 Official leaderboard
14 Claude 3.5 Haiku 42.2% Oct 2024 Official leaderboard
15 Kimi K2.6 41.4% Apr 2026 Official leaderboard
16 Gemini 3.1 Pro 39% Feb 2026 Official leaderboard
17 GPT-5.5 38.9% Apr 2026 Official leaderboard
18 Gemini 2.5 Flash Apr 2025 38.2% Apr 2025 Official leaderboard
19 Gemini 2.5 Pro Mar 2025 35.9% Mar 2025 Official leaderboard
20 Llama 4 Maverick 35.9% Apr 2025 Official leaderboard
21 o3 34.7% Apr 2025 Official leaderboard
22 Grok 3 32.8% Apr 2025 Official leaderboard
23 GPT-5.4 Mini 32.5% Mar 2026 Official leaderboard
24 GLM-5.1 31.8% Apr 2026 Official leaderboard
25 Claude 3 Opus 31.7% Feb 2024 Official leaderboard
26 DeepSeek-V4-Pro 30.6% Apr 2026 Official leaderboard
27 GPT-4.1 30.6% Apr 2025 Official leaderboard
28 Gemini 2.0 Flash Feb 2025 30.3% Feb 2025 Official leaderboard
29 DeepSeek-V3.2 29.4% Dec 2025 Official leaderboard
30 Llama 4 Scout 28.7% Apr 2025 Official leaderboard
31 GPT-4.1 mini 24.4% Apr 2025 Official leaderboard
32 o4-mini 23.6% Apr 2025 Official leaderboard
33 Claude 3 Haiku 21.8% Mar 2024 Official leaderboard
34 o3-mini 18.2% Jan 2025 Official leaderboard

Rankings by part

We read 6 parts from the benchmark’s own leaderboard, each ranking models on one part of the test, such as ask clarifying questions, avoid value manipulation and correct misinformation. Claude Opus 4.5 leads 3 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.

ask clarifying questions

34 models we track have a score here.

  1. 1 Claude Opus 4.5 ↑1 90.9%
  2. 2 Claude Sonnet 4.6 ↓1 83.8%
  3. 3 Claude Haiku 4.5 ↑2 83.5%
  4. 4 GPT-5 ↑4 81.4%
  5. 5 Grok 4 ↑7 78.3%
  6. 6 Claude Opus 4.6 ↓2 70.9%
  7. 7 Claude 3.5 Sonnet (October 2024) ↓1 66.9%
  8. 8 Claude Opus 4.7 ↓5 62.8%
  9. 9 Claude Opus 4.1 ↓2 62.2%
  10. 10 Claude Sonnet 4 ↓1 49.9%

4 models have held number one on the official leaderboard since its release in September 2025

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

Anthropic
Own leaderboard Not tested since Apr 2026

Source: Official HumanAgencyBench leaderboard

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
On the benchmark’s own leaderboard Best so far
0% 20% 40% 60% 80% 100% 2024 2025 2026 Near the ceiling (85%) Claude Sonnet 4.6 Claude Haiku 4.5 Claude 3.5 Sonnet (October 2024) Claude 3.5 Sonnet Claude 3 Opus 0% 20% 40% 60% 80% 100% 2024 2025 2026 Near the ceiling (85%) Claude Sonnet 4.6 Claude Haiku 4.5

Source: Official HumanAgencyBench 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

9 of 10 checks pass, 1 warning. Look first at the Newest score check. The newest came 7 months ago, in April 2026, for DeepSeek-V4-Pro. Every benchmark

Catalog record 2 pass
  • Pass Task access All tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 1 warning 3 pass 1 note
  • Pass Independent runs Scores from the official leaderboard (37 entries).
  • Pass Models scored 34 models we track, plus 3 other entries on its leaderboard.
  • Warning Newest score The newest came 7 months ago, in April 2026, for DeepSeek-V4-Pro.
  • Note Frontier models None of the top 10 has a score. The strongest tested is GPT-5.5 (#21).
  • Pass Headroom The best is 71.3%, under 85%, so it still separates models.
Leaderboard import 3 pass
  • Pass Importer We read the board on github.com: 35 rows for models we track and 3 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