BenchLeader
Menu
Best LLMsChoose a modelCoding plansLeaderboardsModelsBenchmarksProvidersReleases7

MASK leaderboard

As of 6 Nov 2026, the top score on MASK is 96.3% by Claude Opus 4.6 (Anthropic), ahead of Claude Sonnet 4.5 at 96.1%, across 64 model configurations with a published result (Scale AI SEAL, result dated 17 Feb 2026). MASK separates honesty from accuracy: the model's belief is elicited, then it is pressured to say otherwise, and honesty is whether it holds. MASK is shown for reference only and does not count toward the BenchLeader Index.

Key facts (as of 6 Nov 2026)

  • Claude Opus 4.6 holds the top MASK score at 96.3% (Scale AI SEAL, 17 Feb 2026).
  • Claude Sonnet 4.5 is second on MASK at 96.1% (Scale AI SEAL, 2 Oct 2025).
  • MASK: Whether a model states what it believes when pressured to lie. Scale AI (Scale AI SEAL).
  • 64 model configurations have a published MASK result on BenchLeader (BenchLeader, 6 Nov 2026).
  • MASK does not feed the BenchLeader Index; it is shown for reference (BenchLeader methodology).

Whether a model states what it believes when pressured to lie. Scale AI.

Scale AI SEAL marks 37 of these 64 results as possibly contaminated: the model was evaluated after the questions were public, so they may already be in its training data. Those scores are in the table below; this board does not feed the index.

Published by
Scale AI SEAL
Category
Safety & honesty
Index weight
Reference only
Models
64
Data as of
6 Nov 2026

Scale AI SEAL Leaderboards.

What the test looks like

MASK separates honesty from accuracy: the model's belief is elicited, then it is pressured to say otherwise, and honesty is whether it holds.

How it is scored

Honesty rate, published by Scale AI with the Center for AI Safety.

What to keep in mind

A safety and alignment measure; not part of the index.

Top 15
  1. 1Claude Opus 4.6 (none)96.3%
  2. 2Claude Sonnet 4.5 (thinking)96.1%
  3. 3Claude Sonnet 4 (thinking)95.3%
  4. 4Claude Opus 4.1 (thinking)94.2%
  5. 5Claude Opus 4.5 (thinking)92.5%
  6. 6gpt-oss-120b92.0%
  7. 7GPT-5.4 Pro91.7%
  8. 8GPT-5.4 (thinking)89.7%
  9. 9Claude Sonnet 489.3%
  10. 10Claude Opus 4 (thinking)87.9%
  11. 11Claude Opus 4.187.4%
  12. 12Claude Opus 4.587.1%
  13. 13GPT-5.286.7%
  14. 14gpt-oss-20b86.5%
  15. 15Claude Sonnet 4.586.4%
64 of 64
#
1Claude Opus 4.6no reasoning96.3%57.92026-02-17possible
2Claude Sonnet 4.5thinking96.1%52.22025-10-02possible
3Claude Sonnet 4thinking95.3%53.32025-05-24–
4Claude Opus 4.1thinking94.2%57.82025-08-08possible
5Claude Opus 4.5thinking92.5%59.62025-11-26possible
6gpt-oss-120b92.0%45.52025-08-13possible
7GPT-5.4 Pro91.7%63.12026-03-23possible
8GPT-5.4thinking89.7%–2026-03-10possible
9Claude Sonnet 489.3%50.52025-05-23–
10Claude Opus 4thinking87.9%55.22025-05-24–
11Claude Opus 4.187.4%51.42025-08-08possible
12Claude Opus 4.587.1%57.92025-11-26possible
13GPT-5.286.7%57.92025-12-15possible
14gpt-oss-20b86.5%44.02025-08-13possible
15Claude Sonnet 4.586.4%53.72025-10-02possible
16GPT-5.1thinking86.3%54.02025-11-26possible
17GPT-5 Pro86.0%59.92025-11-06possible
18Claude Opus 4.6max85.4%55.82026-02-17possible
19o3high84.5%53.82025-04-25possible
20GPT-5 mini82.6%56.42025-08-22possible
21o3medium82.6%53.32025-04-16possible
22o3-prohigh82.5%53.52025-06-26possible
23Claude 3.7 Sonnetthinking82.1%51.92025-03-05–
24Claude Opus 480.3%52.62025-05-23possible
25GPT-579.3%57.72025-08-13possible
26Claude 3 Opus79.0%39.02025-03-05–
27o4-minihigh78.6%50.32025-04-25possible
28o4-minimedium72.9%51.82025-04-16possible
29Claude 3.5 Sonnet72.3%47.12025-03-05–
30Claude 3.7 Sonnet72.3%49.92025-03-05–
31Kimi K2.570.5%54.22026-02-12possible
32GPT 5.1 Instant63.3%–2025-11-26possible
33o1-pro61.6%–2025-03-05–
34GLM-4.561.5%51.62025-08-13possible
35Llama 3.1 405B61.4%41.92025-03-05–
36GPT-4.1 nano61.4%37.02025-04-14possible
37GLM-4.5-Air60.8%46.72025-08-13possible
38GPT-4o60.1%43.32025-03-05–
39o159.3%53.42025-03-05–
40DeepSeek R157.3%47.32025-03-05–
41GPT-4.556.9%50.82025-03-05–
42Mistral Magistral56.5%–2025-06-26possible
43Qwen3 235B A22B56.4%50.02025-05-01possible
44Gemini 2.5 Pro55.9%54.52025-03-05–
45Llama 3.2 90B Vision54.1%36.12025-03-05–
46DeepSeek R1 052853.0%49.32025-05-29–
47Llama 3.3 70B51.9%40.02025-03-05–
48GPT-4.151.1%48.52025-04-15possible
49GPT-4.1 mini50.0%44.82025-04-14possible
50Llama 4 Maverick49.7%42.12025-03-05–
51o3-minilow49.7%40.32026-03-25–
52Gemini 2.0 Flash Thinkingthinking49.5%45.32025-03-05–
53Gemini 2.5 Flash49.1%52.72026-03-25possible
54Gemini 2.0 Flash49.1%43.52025-03-05–
55o3-minimedium48.9%45.22025-03-05–
56Gemini 2.0 Pro48.7%47.92025-04-01–
57Gemini 3.1 Flash Lite48.4%52.42026-03-23possible
58Mistral Large 2.147.5%42.62025-04-01–
59o3-minihigh46.8%47.72025-04-01–
60Kimi K246.7%51.02025-07-23possible

More safety & honesty benchmarks · All LLM benchmarks

Related benchmarks

Top head-to-heads on MASK

Cite as: BenchLeader, “MASK leaderboard”, data as of 6 Nov 2026, https://www.benchleader.com/benchmarks/scale_mask.

MASK: questions

What does MASK measure?
MASK separates honesty from accuracy: the model's belief is elicited, then it is pressured to say otherwise, and honesty is whether it holds. Scores are reported in percent of tasks solved; higher is better.
What is the top score on MASK?
As of 6 Nov 2026, the highest score on MASK is 96.3% by Claude Opus 4.6 (result dated 17 Feb 2026), ahead of Claude Sonnet 4.5 at 96.1%. Scores come from Scale AI SEAL and are re-read daily.
Which AI model leads MASK?
Claude Opus 4.6 leads MASK with 96.3% as of 6 Nov 2026 (result dated 17 Feb 2026), ahead of Claude Sonnet 4.5 at 96.1%.
How many models have MASK results?
64 model configurations have a MASK result on BenchLeader, all taken from Scale AI SEAL.
Who runs MASK and how often is it updated?
MASK is published by Scale AI SEAL. BenchLeader re-reads the published results once a day and records the date each result was published.
Does MASK count toward the BenchLeader Index?
No. MASK is shown for reference but left out of the composite index.