Benchmarks · /benchmarks/scale-mask

MASK

Name: MASK
Creator: Scale Labs

Scale hidden-goal safety benchmark focused on honesty under conflicting incentives.

Source · Scale Labs
Version · scale-labs snapshot 2026-10-03
Scores · 68

Test details

Visible tradeoffsThis is a rubric-judged signal, so it is more structured than arena taste but still depends on the scoring rubric.

source

Scale Labs

metric

Honesty score (%)

judge

Rubric

direction

higher better

group id

scale_mask_current

domain

Safety

What it measures vs what it misses

✓ Measures

Whether a model stays honest instead of covertly optimizing against the user.

✗ Misses

General capability breadth. Tool-use or retrieval quality.

Why this countsWhether a model stays honest instead of covertly optimizing against the user.Same-test ruleThis percentile only compares models inside the exact benchmark/version group shown here. It is not a universal score.What it missesGeneral capability breadth.

Leaderboard · this benchmark version

#1 · Claude Opus 4.6

SL · Oct 3, 2026

Source label: claude-opus-4-6 (Non-Thinking)

verified runtimeexact aliasBackground only

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 100%
Last updated: recent
Eligibility: historical_model
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: claude-opus-4-6 (Non-Thinking). Collapse policy: highest reported score per canonical model.

96.3%

#2 · claude-sonnet-4-5-20250929-thinking

SL · Oct 3, 2026

verified runtimeexact direct

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 98.5%
Last updated: recent
Eligibility: headline eligible
Identity: exact (1.00)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: claude-sonnet-4-5-20250929-thinking. Collapse policy: highest reported score per canonical model.

96.1%

#3 · Claude Sonnet 4 (Thinking)

SL · Oct 3, 2026

Source label: Claude Sonnet 4 (Thinking)

verified runtimeexact alias

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 97%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: Claude Sonnet 4 (Thinking). Collapse policy: highest reported score per canonical model.

95.3%

#4 · claude-opus-4-1-20250805-thinking

SL · Oct 3, 2026

verified runtimeexact direct

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 95.5%
Last updated: recent
Eligibility: headline eligible
Identity: exact (1.00)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: claude-opus-4-1-20250805-thinking. Collapse policy: highest reported score per canonical model.

94.2%

#5 · claude-opus-4-5-20251101-thinking

SL · Oct 3, 2026

verified runtimeexact direct

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 94%
Last updated: recent
Eligibility: headline eligible
Identity: exact (1.00)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: claude-opus-4-5-20251101-thinking. Collapse policy: highest reported score per canonical model.

92.5%

#6 · GPT-OSS 120B

SL · Oct 3, 2026

verified runtimeexact directBackground only

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 92.5%
Last updated: recent
Eligibility: historical_model
Identity: exact (1.00)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: gpt-oss-120b. Collapse policy: highest reported score per canonical model.

92%

#7 · GPT-5.4

SL · Oct 3, 2026

Source label: gpt-5.4-pro-2026-03-05

verified runtimeexact alias

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 91%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: gpt-5.4-pro-2026-03-05. Collapse policy: highest reported score per canonical model.

91.7%

#8 · Claude Sonnet 4

SL · Oct 3, 2026

Source label: Claude Sonnet 4

verified runtimeexact aliasBackground only

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 89.6%
Last updated: recent
Eligibility: historical_model
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: Claude Sonnet 4. Collapse policy: highest reported score per canonical model.

89.3%

#9 · Claude Sonnet 4.6

SL · Oct 3, 2026

Source label: Claude Sonnet 4

backfilledproxy backfilledBackground only

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 89.6%
Last updated: recent
Eligibility: Fallback benchmark identity is visible for context but excluded from default ranking.
Identity: benchmark proxy (0.58)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: Claude Sonnet 4. Collapse policy: highest reported score per canonical model. Backfilled from Claude Sonnet 4 via approved benchmark identity mapping map-claude-sonnet-4-6-to-4.

89.3%

#10 · Claude Opus 4 (Thinking)

SL · Oct 3, 2026

Source label: Claude Opus 4 (Thinking)

verified runtimeexact alias

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 86.6%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: Claude Opus 4 (Thinking). Collapse policy: highest reported score per canonical model.

87.9%

#11 · Claude Opus 4.1

SL · Oct 3, 2026

Source label: claude-opus-4-1-20250805

verified runtimeexact aliasBackground only

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 85.1%
Last updated: recent
Eligibility: historical_model
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: claude-opus-4-1-20250805. Collapse policy: highest reported score per canonical model.

87.4%

#12 · Claude Opus 4.7

SL · Oct 3, 2026

Source label: claude-opus-4-1-20250805

backfilledproxy backfilledBackground only

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 85.1%
Last updated: recent
Eligibility: Fallback benchmark identity is visible for context but excluded from default ranking.
Identity: benchmark proxy (0.58)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: claude-opus-4-1-20250805. Collapse policy: highest reported score per canonical model. Backfilled from Claude Opus 4.1 via approved benchmark identity mapping map-claude-opus-4-7-to-4-1.

87.4%

#13 · Claude Opus 4.5

SL · Oct 3, 2026

Source label: claude-opus-4-5-20251101

verified runtimeexact aliasBackground only

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 82.1%
Last updated: recent
Eligibility: historical_model
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: claude-opus-4-5-20251101. Collapse policy: highest reported score per canonical model.

87.1%

#14 · gpt-5.2-2025-12-11

SL · Oct 3, 2026

Source label: gpt-5.2-2025-12-11

verified runtimeexact alias

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 80.6%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: gpt-5.2-2025-12-11. Collapse policy: highest reported score per canonical model.

86.7%

#15 · GPT-OSS 20B

SL · Oct 3, 2026

verified runtimeexact directBackground only

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 79.1%
Last updated: recent
Eligibility: historical_model
Identity: exact (1.00)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: gpt-oss-20b. Collapse policy: highest reported score per canonical model.

86.5%

#16 · Claude Sonnet 4.5

SL · Oct 3, 2026

Source label: claude-sonnet-4-5-20250929

verified runtimeexact aliasBackground only

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 77.6%
Last updated: recent
Eligibility: historical_model
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: claude-sonnet-4-5-20250929. Collapse policy: highest reported score per canonical model.

86.4%

#17 · GPT-5.1

SL · Oct 3, 2026

Source label: gpt-5.1-thinking

verified runtimeexact aliasBackground only

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 76.1%
Last updated: recent
Eligibility: historical_model
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: gpt-5.1-thinking. Collapse policy: highest reported score per canonical model.

86.3%

#18 · GPT-5

SL · Oct 3, 2026

Source label: gpt-5-pro-2025-10-06

verified runtimeexact alias

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 74.6%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: gpt-5-pro-2025-10-06. Collapse policy: highest reported score per canonical model.

86%

#19 · GPT-5.4 mini

SL · Oct 3, 2026

Source label: gpt-5-pro-2025-10-06

backfilledproxy backfilledBackground only

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 74.6%
Last updated: recent
Eligibility: Fallback benchmark identity is visible for context but excluded from default ranking.
Identity: benchmark proxy (0.58)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: gpt-5-pro-2025-10-06. Collapse policy: highest reported score per canonical model. Backfilled from GPT-5 via approved benchmark identity mapping map-gpt-5-4-mini-to-gpt-5.

86%

#20 · GPT-5.4 nano

SL · Oct 3, 2026

Source label: gpt-5-pro-2025-10-06

backfilledproxy backfilledBackground only

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 74.6%
Last updated: recent
Eligibility: Fallback benchmark identity is visible for context but excluded from default ranking.
Identity: benchmark proxy (0.58)

86%

#21 · o3 (high) (April 2025)

SL · Oct 3, 2026

Source label: o3 (high) (April 2025)

verified runtimeexact alias

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 70.1%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: o3 (high) (April 2025). Collapse policy: highest reported score per canonical model.

84.5%

#22 · o3 (medium) (April 2025)

SL · Oct 3, 2026

Source label: o3 (medium) (April 2025)

verified runtimeexact alias

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 68.7%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: o3 (medium) (April 2025). Collapse policy: highest reported score per canonical model.

82.6%

#23 · gpt-5-mini-2025-08-07

SL · Oct 3, 2026

verified runtimeexact direct

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 68.7%
Last updated: recent
Eligibility: headline eligible
Identity: exact (1.00)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: gpt-5-mini-2025-08-07. Collapse policy: highest reported score per canonical model.

82.6%

#24 · o3 Pro (high) (June 2025)

SL · Oct 3, 2026

Source label: o3 Pro (high) (June 2025)

verified runtimeexact alias

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 65.7%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: o3 Pro (high) (June 2025). Collapse policy: highest reported score per canonical model.

82.5%

#25 · Claude 3.7 Sonnet (Thinking) (February 2025)

SL · Oct 3, 2026

Source label: Claude 3.7 Sonnet (Thinking) (February 2025)

verified runtimeexact alias

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 64.2%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: Claude 3.7 Sonnet (Thinking) (February 2025). Collapse policy: highest reported score per canonical model.

82.1%

#26 · Claude Opus 4

SL · Oct 3, 2026

Source label: Claude Opus 4

verified runtimeexact aliasBackground only

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 62.7%
Last updated: recent
Eligibility: historical_model
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: Claude Opus 4. Collapse policy: highest reported score per canonical model.

80.3%

#27 · Claude Opus 3

SL · Oct 3, 2026

Source label: Claude 3 Opus

verified runtimeexact aliasBackground only

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 61.2%
Last updated: recent
Eligibility: historical_model
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: Claude 3 Opus. Collapse policy: highest reported score per canonical model.

79%

#28 · o4-mini (high) (April 2025)

SL · Oct 3, 2026

Source label: o4-mini (high) (April 2025)

verified runtimeexact alias

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 59.7%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: o4-mini (high) (April 2025). Collapse policy: highest reported score per canonical model.

78.6%

#29 · o4-mini (medium) (April 2025)

SL · Oct 3, 2026

Source label: o4-mini (medium) (April 2025)

verified runtimeexact alias

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 58.2%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: o4-mini (medium) (April 2025). Collapse policy: highest reported score per canonical model.

72.9%

#30 · Claude 3.5 Sonnet (October 2024)

SL · Oct 3, 2026

Source label: Claude 3.5 Sonnet (October 2024)

verified runtimeexact alias

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 56.7%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: Claude 3.5 Sonnet (October 2024). Collapse policy: highest reported score per canonical model.

72.3%

#31 · Claude 3.7 Sonnet (February 2025)

SL · Oct 3, 2026

Source label: Claude 3.7 Sonnet (February 2025)

verified runtimeexact alias

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 55.2%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: Claude 3.7 Sonnet (February 2025). Collapse policy: highest reported score per canonical model.

72.3%

#32 · Kimi K2.5 (Reasoning)

SL · Oct 3, 2026

Source label: kimi-k2.5

verified runtimeexact aliasBackground only

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 53.7%
Last updated: recent
Eligibility: benchmark_derived_model
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: kimi-k2.5. Collapse policy: highest reported score per canonical model.

70.5%

#33 · gpt-5.1-instant

SL · Oct 3, 2026

Source label: gpt-5.1-instant

verified runtimeexact alias

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 52.2%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: gpt-5.1-instant. Collapse policy: highest reported score per canonical model.

63.3%

#34 · o1-pro

SL · Oct 3, 2026

Source label: o1-Pro

verified runtimeexact alias

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 50.7%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: o1-Pro. Collapse policy: highest reported score per canonical model.

61.6%

#35 · glm-4p5

SL · Oct 3, 2026

verified runtimeexact direct

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 49.3%
Last updated: recent
Eligibility: headline eligible
Identity: exact (1.00)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: glm-4p5. Collapse policy: highest reported score per canonical model.

61.5%

#36 · GPT-4.1 nano

SL · Oct 3, 2026

Source label: GPT-4.1 nano

verified runtimeexact aliasBackground only

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 47.8%
Last updated: recent
Eligibility: historical_model
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: GPT-4.1 nano. Collapse policy: highest reported score per canonical model.

61.4%

#37 · Llama 3.1 405B Instruct

SL · Oct 3, 2026

Source label: Llama 3.1 405B Instruct

verified runtimeexact alias

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 47.8%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: Llama 3.1 405B Instruct. Collapse policy: highest reported score per canonical model.

61.4%

#38 · glm-4p5-air

SL · Oct 3, 2026

verified runtimeexact direct

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 44.8%
Last updated: recent
Eligibility: headline eligible
Identity: exact (1.00)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: glm-4p5-air. Collapse policy: highest reported score per canonical model.

60.8%

#39 · gpt 4o (November 2024)

SL · Oct 3, 2026

Source label: gpt 4o (November 2024)

verified runtimeexact alias

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 43.3%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: gpt 4o (November 2024). Collapse policy: highest reported score per canonical model.

60.1%

#40 · o1 (December 2024)

SL · Oct 3, 2026

Source label: o1 (December 2024)

verified runtimeexact alias

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 41.8%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: o1 (December 2024). Collapse policy: highest reported score per canonical model.

59.3%

#41 · DeepSeek R1 0528 (May '25)

SL · Oct 3, 2026

Source label: Deepseek R1 (Jan 2025)

verified runtimeexact alias

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 40.3%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: Deepseek R1 (Jan 2025). Collapse policy: highest reported score per canonical model.

57.3%

#42 · GPT-4.5 Preview

SL · Oct 3, 2026

Source label: GPT 4.5 Preview

verified runtimeexact aliasBackground only

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 38.8%
Last updated: recent
Eligibility: historical_model
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: GPT 4.5 Preview. Collapse policy: highest reported score per canonical model.

56.9%

#43 · Mistral Magistral

SL · Oct 3, 2026

Source label: Mistral Magistral

verified runtimeexact alias

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 37.3%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: Mistral Magistral. Collapse policy: highest reported score per canonical model.

56.5%

#44 · Qwen3 235B A22B

SL · Oct 3, 2026

Source label: Qwen3-235B-A22B

verified runtimeexact alias

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 35.8%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: Qwen3-235B-A22B. Collapse policy: highest reported score per canonical model.

56.4%

#45 · Gemini 2.5 Pro Experimental (March 2025)

SL · Oct 3, 2026

Source label: Gemini 2.5 Pro Experimental (March 2025)

verified runtimeexact aliasBackground only

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 34.3%
Last updated: recent
Eligibility: preview_model
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: Gemini 2.5 Pro Experimental (March 2025). Collapse policy: highest reported score per canonical model.

55.9%

#46 · Gemini 2.5 Pro

SL · Oct 3, 2026

Source label: gemini-2.5-pro-preview-06-05

verified runtimeexact alias

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 32.8%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: gemini-2.5-pro-preview-06-05. Collapse policy: highest reported score per canonical model.

55.7%

#47 · Llama 3.2 90B Vision Instruct

SL · Oct 3, 2026

Source label: Llama 3.2 90B Vision Instruct

verified runtimeexact alias

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 31.3%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: Llama 3.2 90B Vision Instruct. Collapse policy: highest reported score per canonical model.

54.1%

#48 · Gemini 2.5 Pro Preview (May 06 2025)

SL · Oct 3, 2026

Source label: Gemini 2.5 Pro Preview (May 06 2025)

verified runtimeexact aliasBackground only

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 29.9%
Last updated: recent
Eligibility: preview_model
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: Gemini 2.5 Pro Preview (May 06 2025). Collapse policy: highest reported score per canonical model.

53.1%

#49 · deepseek-r1-0528

SL · Oct 3, 2026

Source label: DeepSeek-R1-0528

verified runtimeexact alias

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 28.4%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: DeepSeek-R1-0528. Collapse policy: highest reported score per canonical model.

53%

#50 · llama-3.3-70b-instruct

SL · Oct 3, 2026

Source label: Llama 3.3 70B Instruct

verified runtimeexact alias

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 26.9%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: Llama 3.3 70B Instruct. Collapse policy: highest reported score per canonical model.

51.9%

#51 · GPT-4.1

SL · Oct 3, 2026

Source label: GPT-4.1

verified runtimeexact aliasBackground only

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 25.4%
Last updated: recent
Eligibility: historical_model
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: GPT-4.1. Collapse policy: highest reported score per canonical model.

51.1%

#52 · GPT-4.1 mini

SL · Oct 3, 2026

Source label: GPT-4.1 mini

verified runtimeexact aliasBackground only

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 23.9%
Last updated: recent
Eligibility: historical_model
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: GPT-4.1 mini. Collapse policy: highest reported score per canonical model.

50%

#53 · Llama 4 Maverick

SL · Oct 3, 2026

Source label: Llama 4 Maverick

verified runtimeexact alias

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 22.4%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: Llama 4 Maverick. Collapse policy: highest reported score per canonical model.

49.7%

#54 · o3 mini (Low)

SL · Oct 3, 2026

Source label: o3 mini (Low)

verified runtimeexact alias

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 22.4%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: o3 mini (Low). Collapse policy: highest reported score per canonical model.

49.7%

#55 · Gemini 2.0 Flash Thinking (January 2025)

SL · Oct 3, 2026

Source label: Gemini 2.0 Flash Thinking (January 2025)

verified runtimeexact alias

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 19.4%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: Gemini 2.0 Flash Thinking (January 2025). Collapse policy: highest reported score per canonical model.

49.5%

#56 · Gemini 2.5 Flash Preview (May 2025)

SL · Oct 3, 2026

Source label: Gemini 2.5 Flash Preview (May 2025)

verified runtimeexact aliasBackground only

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 17.9%
Last updated: recent
Eligibility: preview_model
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: Gemini 2.5 Flash Preview (May 2025). Collapse policy: highest reported score per canonical model.

49.1%

#57 · Gemini 2.0 Flash

SL · Oct 3, 2026

Source label: Gemini 2.0 Flash

verified runtimeexact aliasBackground only

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 16.4%
Last updated: recent
Eligibility: historical_model
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: Gemini 2.0 Flash. Collapse policy: highest reported score per canonical model.

49.1%

#58 · o3 mini (Medium)

SL · Oct 3, 2026

Source label: o3 mini (Medium)

verified runtimeexact alias

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 14.9%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: o3 mini (Medium). Collapse policy: highest reported score per canonical model.

48.9%

#59 · Gemini 2.0 Pro Experimental (February 2025)

SL · Oct 3, 2026

Source label: Gemini 2.0 Pro Experimental (February 2025)

verified runtimeexact aliasBackground only

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 13.4%
Last updated: recent
Eligibility: preview_model
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: Gemini 2.0 Pro Experimental (February 2025). Collapse policy: highest reported score per canonical model.

48.7%

#60 · Gemini 3.1 Flash-Lite Preview

SL · Oct 3, 2026

Source label: gemini-3.1-flash-lite-preview

verified runtimeexact aliasBackground only

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 11.9%
Last updated: recent
Eligibility: preview_model
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: gemini-3.1-flash-lite-preview. Collapse policy: highest reported score per canonical model.

48.4%

#61 · mistral-large-2411

SL · Oct 3, 2026

Source label: Mistral Large 2411

verified runtimeexact alias

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 10.4%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: Mistral Large 2411. Collapse policy: highest reported score per canonical model.

47.5%

#62 · o3 mini

SL · Oct 3, 2026

Source label: o3 mini (High)

verified runtimeexact aliasBackground only

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 9%
Last updated: recent
Eligibility: historical_model
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: o3 mini (High). Collapse policy: highest reported score per canonical model.

46.8%

#63 · kimi-k2-instruct

SL · Oct 3, 2026

verified runtimeexact directBackground only

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 7.5%
Last updated: recent
Eligibility: benchmark_derived_model
Identity: exact (1.00)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: kimi-k2-instruct. Collapse policy: highest reported score per canonical model.

46.7%

#64 · deepseek-v3.1

SL · Oct 3, 2026

Source label: DeepSeek-V3.1

verified runtimeexact alias

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 6%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: DeepSeek-V3.1. Collapse policy: highest reported score per canonical model.

46.3%

#65 · Deepseek V3 (March 2025)

SL · Oct 3, 2026

Source label: Deepseek V3 (March 2025)

verified runtimeexact aliasBackground only

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 4.5%
Last updated: recent
Eligibility: preview_model
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: Deepseek V3 (March 2025). Collapse policy: highest reported score per canonical model.

44.5%

#66 · Gemini 3 Pro Preview

SL · Oct 3, 2026

verified runtimeexact directBackground only

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 3%
Last updated: recent
Eligibility: preview_model
Identity: exact (1.00)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: gemini-3-pro-preview. Collapse policy: highest reported score per canonical model.

42.6%

#67 · Mistral Medium 3

SL · Oct 3, 2026

Source label: Mistral Medium 3

verified runtimeexact aliasBackground only

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 3%
Last updated: recent
Eligibility: historical_model
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: Mistral Medium 3. Collapse policy: highest reported score per canonical model.

42.6%

#68 · Gemini 3.1 Pro Preview

SL · Oct 3, 2026

Source label: gemini-3.1-pro-preview

verified runtimeexact aliasBackground only

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 0%
Last updated: recent
Eligibility: preview_model
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: gemini-3.1-pro-preview. Collapse policy: highest reported score per canonical model.

42.4%

Benchmarks · /benchmarks/scale-mask

MASK

Scale hidden-goal safety benchmark focused on honesty under conflicting incentives.

Source · Scale Labs
Version · scale-labs snapshot 2026-10-03
Scores · 68

Test details

Visible tradeoffsThis is a rubric-judged signal, so it is more structured than arena taste but still depends on the scoring rubric.

source

Scale Labs

metric

Honesty score (%)

judge

Rubric

direction

higher better

group id

scale_mask_current

domain

Safety

What it measures vs what it misses

✓ Measures

Whether a model stays honest instead of covertly optimizing against the user.

✗ Misses

General capability breadth. Tool-use or retrieval quality.

Leaderboard · this benchmark version

#1 · Claude Opus 4.6

SL · Oct 3, 2026

Source label: claude-opus-4-6 (Non-Thinking)

verified runtimeexact aliasBackground only

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 100%
Last updated: recent
Eligibility: historical_model
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: claude-opus-4-6 (Non-Thinking). Collapse policy: highest reported score per canonical model.

96.3%

#2 · claude-sonnet-4-5-20250929-thinking

SL · Oct 3, 2026

verified runtimeexact direct

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 98.5%
Last updated: recent
Eligibility: headline eligible
Identity: exact (1.00)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: claude-sonnet-4-5-20250929-thinking. Collapse policy: highest reported score per canonical model.

96.1%

#3 · Claude Sonnet 4 (Thinking)

SL · Oct 3, 2026

Source label: Claude Sonnet 4 (Thinking)

verified runtimeexact alias

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 97%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: Claude Sonnet 4 (Thinking). Collapse policy: highest reported score per canonical model.

95.3%

#4 · claude-opus-4-1-20250805-thinking

SL · Oct 3, 2026

verified runtimeexact direct

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 95.5%
Last updated: recent
Eligibility: headline eligible
Identity: exact (1.00)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: claude-opus-4-1-20250805-thinking. Collapse policy: highest reported score per canonical model.

94.2%

#5 · claude-opus-4-5-20251101-thinking

SL · Oct 3, 2026

verified runtimeexact direct

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 94%
Last updated: recent
Eligibility: headline eligible
Identity: exact (1.00)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: claude-opus-4-5-20251101-thinking. Collapse policy: highest reported score per canonical model.

92.5%

#6 · GPT-OSS 120B

SL · Oct 3, 2026

verified runtimeexact directBackground only

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 92.5%
Last updated: recent
Eligibility: historical_model
Identity: exact (1.00)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: gpt-oss-120b. Collapse policy: highest reported score per canonical model.

92%

#7 · GPT-5.4

SL · Oct 3, 2026

Source label: gpt-5.4-pro-2026-03-05

verified runtimeexact alias

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 91%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: gpt-5.4-pro-2026-03-05. Collapse policy: highest reported score per canonical model.

91.7%

#8 · Claude Sonnet 4

SL · Oct 3, 2026

Source label: Claude Sonnet 4

verified runtimeexact aliasBackground only

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 89.6%
Last updated: recent
Eligibility: historical_model
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: Claude Sonnet 4. Collapse policy: highest reported score per canonical model.

89.3%

#9 · Claude Sonnet 4.6

SL · Oct 3, 2026

Source label: Claude Sonnet 4

backfilledproxy backfilledBackground only

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 89.6%
Last updated: recent
Eligibility: Fallback benchmark identity is visible for context but excluded from default ranking.
Identity: benchmark proxy (0.58)

89.3%

#10 · Claude Opus 4 (Thinking)

SL · Oct 3, 2026

Source label: Claude Opus 4 (Thinking)

verified runtimeexact alias

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 86.6%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: Claude Opus 4 (Thinking). Collapse policy: highest reported score per canonical model.

87.9%

#11 · Claude Opus 4.1

SL · Oct 3, 2026

Source label: claude-opus-4-1-20250805

verified runtimeexact aliasBackground only

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 85.1%
Last updated: recent
Eligibility: historical_model
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: claude-opus-4-1-20250805. Collapse policy: highest reported score per canonical model.

87.4%

#12 · Claude Opus 4.7

SL · Oct 3, 2026

Source label: claude-opus-4-1-20250805

backfilledproxy backfilledBackground only

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 85.1%
Last updated: recent
Eligibility: Fallback benchmark identity is visible for context but excluded from default ranking.
Identity: benchmark proxy (0.58)

87.4%

#13 · Claude Opus 4.5

SL · Oct 3, 2026

Source label: claude-opus-4-5-20251101

verified runtimeexact aliasBackground only

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 82.1%
Last updated: recent
Eligibility: historical_model
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: claude-opus-4-5-20251101. Collapse policy: highest reported score per canonical model.

87.1%

#14 · gpt-5.2-2025-12-11

SL · Oct 3, 2026

Source label: gpt-5.2-2025-12-11

verified runtimeexact alias

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 80.6%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: gpt-5.2-2025-12-11. Collapse policy: highest reported score per canonical model.

86.7%

#15 · GPT-OSS 20B

SL · Oct 3, 2026

verified runtimeexact directBackground only

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 79.1%
Last updated: recent
Eligibility: historical_model
Identity: exact (1.00)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: gpt-oss-20b. Collapse policy: highest reported score per canonical model.

86.5%

#16 · Claude Sonnet 4.5

SL · Oct 3, 2026

Source label: claude-sonnet-4-5-20250929

verified runtimeexact aliasBackground only

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 77.6%
Last updated: recent
Eligibility: historical_model
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: claude-sonnet-4-5-20250929. Collapse policy: highest reported score per canonical model.

86.4%

#17 · GPT-5.1

SL · Oct 3, 2026

Source label: gpt-5.1-thinking

verified runtimeexact aliasBackground only

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 76.1%
Last updated: recent
Eligibility: historical_model
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: gpt-5.1-thinking. Collapse policy: highest reported score per canonical model.

86.3%

#18 · GPT-5

SL · Oct 3, 2026

Source label: gpt-5-pro-2025-10-06

verified runtimeexact alias

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 74.6%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: gpt-5-pro-2025-10-06. Collapse policy: highest reported score per canonical model.

86%

#19 · GPT-5.4 mini

SL · Oct 3, 2026

Source label: gpt-5-pro-2025-10-06

backfilledproxy backfilledBackground only

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 74.6%
Last updated: recent
Eligibility: Fallback benchmark identity is visible for context but excluded from default ranking.
Identity: benchmark proxy (0.58)

86%

#20 · GPT-5.4 nano

SL · Oct 3, 2026

Source label: gpt-5-pro-2025-10-06

backfilledproxy backfilledBackground only

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 74.6%
Last updated: recent
Eligibility: Fallback benchmark identity is visible for context but excluded from default ranking.
Identity: benchmark proxy (0.58)

86%

#21 · o3 (high) (April 2025)

SL · Oct 3, 2026

Source label: o3 (high) (April 2025)

verified runtimeexact alias

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 70.1%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: o3 (high) (April 2025). Collapse policy: highest reported score per canonical model.

84.5%

#22 · o3 (medium) (April 2025)

SL · Oct 3, 2026

Source label: o3 (medium) (April 2025)

verified runtimeexact alias

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 68.7%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: o3 (medium) (April 2025). Collapse policy: highest reported score per canonical model.

82.6%

#23 · gpt-5-mini-2025-08-07

SL · Oct 3, 2026

verified runtimeexact direct

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 68.7%
Last updated: recent
Eligibility: headline eligible
Identity: exact (1.00)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: gpt-5-mini-2025-08-07. Collapse policy: highest reported score per canonical model.

82.6%

#24 · o3 Pro (high) (June 2025)

SL · Oct 3, 2026

Source label: o3 Pro (high) (June 2025)

verified runtimeexact alias

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 65.7%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: o3 Pro (high) (June 2025). Collapse policy: highest reported score per canonical model.

82.5%

#25 · Claude 3.7 Sonnet (Thinking) (February 2025)

SL · Oct 3, 2026

Source label: Claude 3.7 Sonnet (Thinking) (February 2025)

verified runtimeexact alias

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 64.2%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: Claude 3.7 Sonnet (Thinking) (February 2025). Collapse policy: highest reported score per canonical model.

82.1%

#26 · Claude Opus 4

SL · Oct 3, 2026

Source label: Claude Opus 4

verified runtimeexact aliasBackground only

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 62.7%
Last updated: recent
Eligibility: historical_model
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: Claude Opus 4. Collapse policy: highest reported score per canonical model.

80.3%

#27 · Claude Opus 3

SL · Oct 3, 2026

Source label: Claude 3 Opus

verified runtimeexact aliasBackground only

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 61.2%
Last updated: recent
Eligibility: historical_model
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: Claude 3 Opus. Collapse policy: highest reported score per canonical model.

79%

#28 · o4-mini (high) (April 2025)

SL · Oct 3, 2026

Source label: o4-mini (high) (April 2025)

verified runtimeexact alias

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 59.7%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: o4-mini (high) (April 2025). Collapse policy: highest reported score per canonical model.

78.6%

#29 · o4-mini (medium) (April 2025)

SL · Oct 3, 2026

Source label: o4-mini (medium) (April 2025)

verified runtimeexact alias

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 58.2%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: o4-mini (medium) (April 2025). Collapse policy: highest reported score per canonical model.

72.9%

#30 · Claude 3.5 Sonnet (October 2024)

SL · Oct 3, 2026

Source label: Claude 3.5 Sonnet (October 2024)

verified runtimeexact alias

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 56.7%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: Claude 3.5 Sonnet (October 2024). Collapse policy: highest reported score per canonical model.

72.3%

#31 · Claude 3.7 Sonnet (February 2025)

SL · Oct 3, 2026

Source label: Claude 3.7 Sonnet (February 2025)

verified runtimeexact alias

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 55.2%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: Claude 3.7 Sonnet (February 2025). Collapse policy: highest reported score per canonical model.

72.3%

#32 · Kimi K2.5 (Reasoning)

SL · Oct 3, 2026

Source label: kimi-k2.5

verified runtimeexact aliasBackground only

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 53.7%
Last updated: recent
Eligibility: benchmark_derived_model
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: kimi-k2.5. Collapse policy: highest reported score per canonical model.

70.5%

#33 · gpt-5.1-instant

SL · Oct 3, 2026

Source label: gpt-5.1-instant

verified runtimeexact alias

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 52.2%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: gpt-5.1-instant. Collapse policy: highest reported score per canonical model.

63.3%

#34 · o1-pro

SL · Oct 3, 2026

Source label: o1-Pro

verified runtimeexact alias

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 50.7%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: o1-Pro. Collapse policy: highest reported score per canonical model.

61.6%

#35 · glm-4p5

SL · Oct 3, 2026

verified runtimeexact direct

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 49.3%
Last updated: recent
Eligibility: headline eligible
Identity: exact (1.00)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: glm-4p5. Collapse policy: highest reported score per canonical model.

61.5%

#36 · GPT-4.1 nano

SL · Oct 3, 2026

Source label: GPT-4.1 nano

verified runtimeexact aliasBackground only

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 47.8%
Last updated: recent
Eligibility: historical_model
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: GPT-4.1 nano. Collapse policy: highest reported score per canonical model.

61.4%

#37 · Llama 3.1 405B Instruct

SL · Oct 3, 2026

Source label: Llama 3.1 405B Instruct

verified runtimeexact alias

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 47.8%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: Llama 3.1 405B Instruct. Collapse policy: highest reported score per canonical model.

61.4%

#38 · glm-4p5-air

SL · Oct 3, 2026

verified runtimeexact direct

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 44.8%
Last updated: recent
Eligibility: headline eligible
Identity: exact (1.00)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: glm-4p5-air. Collapse policy: highest reported score per canonical model.

60.8%

#39 · gpt 4o (November 2024)

SL · Oct 3, 2026

Source label: gpt 4o (November 2024)

verified runtimeexact alias

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 43.3%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: gpt 4o (November 2024). Collapse policy: highest reported score per canonical model.

60.1%

#40 · o1 (December 2024)

SL · Oct 3, 2026

Source label: o1 (December 2024)

verified runtimeexact alias

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 41.8%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: o1 (December 2024). Collapse policy: highest reported score per canonical model.

59.3%

#41 · DeepSeek R1 0528 (May '25)

SL · Oct 3, 2026

Source label: Deepseek R1 (Jan 2025)

verified runtimeexact alias

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 40.3%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: Deepseek R1 (Jan 2025). Collapse policy: highest reported score per canonical model.

57.3%

#42 · GPT-4.5 Preview

SL · Oct 3, 2026

Source label: GPT 4.5 Preview

verified runtimeexact aliasBackground only

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 38.8%
Last updated: recent
Eligibility: historical_model
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: GPT 4.5 Preview. Collapse policy: highest reported score per canonical model.

56.9%

#43 · Mistral Magistral

SL · Oct 3, 2026

Source label: Mistral Magistral

verified runtimeexact alias

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 37.3%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: Mistral Magistral. Collapse policy: highest reported score per canonical model.

56.5%

#44 · Qwen3 235B A22B

SL · Oct 3, 2026

Source label: Qwen3-235B-A22B

verified runtimeexact alias

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 35.8%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: Qwen3-235B-A22B. Collapse policy: highest reported score per canonical model.

56.4%

#45 · Gemini 2.5 Pro Experimental (March 2025)

SL · Oct 3, 2026

Source label: Gemini 2.5 Pro Experimental (March 2025)

verified runtimeexact aliasBackground only

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 34.3%
Last updated: recent
Eligibility: preview_model
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: Gemini 2.5 Pro Experimental (March 2025). Collapse policy: highest reported score per canonical model.

55.9%

#46 · Gemini 2.5 Pro

SL · Oct 3, 2026

Source label: gemini-2.5-pro-preview-06-05

verified runtimeexact alias

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 32.8%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: gemini-2.5-pro-preview-06-05. Collapse policy: highest reported score per canonical model.

55.7%

#47 · Llama 3.2 90B Vision Instruct

SL · Oct 3, 2026

Source label: Llama 3.2 90B Vision Instruct

verified runtimeexact alias

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 31.3%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: Llama 3.2 90B Vision Instruct. Collapse policy: highest reported score per canonical model.

54.1%

#48 · Gemini 2.5 Pro Preview (May 06 2025)

SL · Oct 3, 2026

Source label: Gemini 2.5 Pro Preview (May 06 2025)

verified runtimeexact aliasBackground only

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 29.9%
Last updated: recent
Eligibility: preview_model
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: Gemini 2.5 Pro Preview (May 06 2025). Collapse policy: highest reported score per canonical model.

53.1%

#49 · deepseek-r1-0528

SL · Oct 3, 2026

Source label: DeepSeek-R1-0528

verified runtimeexact alias

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 28.4%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: DeepSeek-R1-0528. Collapse policy: highest reported score per canonical model.

53%

#50 · llama-3.3-70b-instruct

SL · Oct 3, 2026

Source label: Llama 3.3 70B Instruct

verified runtimeexact alias

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 26.9%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: Llama 3.3 70B Instruct. Collapse policy: highest reported score per canonical model.

51.9%

#51 · GPT-4.1

SL · Oct 3, 2026

Source label: GPT-4.1

verified runtimeexact aliasBackground only

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 25.4%
Last updated: recent
Eligibility: historical_model
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: GPT-4.1. Collapse policy: highest reported score per canonical model.

51.1%

#52 · GPT-4.1 mini

SL · Oct 3, 2026

Source label: GPT-4.1 mini

verified runtimeexact aliasBackground only

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 23.9%
Last updated: recent
Eligibility: historical_model
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: GPT-4.1 mini. Collapse policy: highest reported score per canonical model.

50%

#53 · Llama 4 Maverick

SL · Oct 3, 2026

Source label: Llama 4 Maverick

verified runtimeexact alias

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 22.4%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: Llama 4 Maverick. Collapse policy: highest reported score per canonical model.

49.7%

#54 · o3 mini (Low)

SL · Oct 3, 2026

Source label: o3 mini (Low)

verified runtimeexact alias

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 22.4%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: o3 mini (Low). Collapse policy: highest reported score per canonical model.

49.7%

#55 · Gemini 2.0 Flash Thinking (January 2025)

SL · Oct 3, 2026

Source label: Gemini 2.0 Flash Thinking (January 2025)

verified runtimeexact alias

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 19.4%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: Gemini 2.0 Flash Thinking (January 2025). Collapse policy: highest reported score per canonical model.

49.5%

#56 · Gemini 2.5 Flash Preview (May 2025)

SL · Oct 3, 2026

Source label: Gemini 2.5 Flash Preview (May 2025)

verified runtimeexact aliasBackground only

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 17.9%
Last updated: recent
Eligibility: preview_model
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: Gemini 2.5 Flash Preview (May 2025). Collapse policy: highest reported score per canonical model.

49.1%

#57 · Gemini 2.0 Flash

SL · Oct 3, 2026

Source label: Gemini 2.0 Flash

verified runtimeexact aliasBackground only

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 16.4%
Last updated: recent
Eligibility: historical_model
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: Gemini 2.0 Flash. Collapse policy: highest reported score per canonical model.

49.1%

#58 · o3 mini (Medium)

SL · Oct 3, 2026

Source label: o3 mini (Medium)

verified runtimeexact alias

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 14.9%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: o3 mini (Medium). Collapse policy: highest reported score per canonical model.

48.9%

#59 · Gemini 2.0 Pro Experimental (February 2025)

SL · Oct 3, 2026

Source label: Gemini 2.0 Pro Experimental (February 2025)

verified runtimeexact aliasBackground only

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 13.4%
Last updated: recent
Eligibility: preview_model
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: Gemini 2.0 Pro Experimental (February 2025). Collapse policy: highest reported score per canonical model.

48.7%

#60 · Gemini 3.1 Flash-Lite Preview

SL · Oct 3, 2026

Source label: gemini-3.1-flash-lite-preview

verified runtimeexact aliasBackground only

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 11.9%
Last updated: recent
Eligibility: preview_model
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: gemini-3.1-flash-lite-preview. Collapse policy: highest reported score per canonical model.

48.4%

#61 · mistral-large-2411

SL · Oct 3, 2026

Source label: Mistral Large 2411

verified runtimeexact alias

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 10.4%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: Mistral Large 2411. Collapse policy: highest reported score per canonical model.

47.5%

#62 · o3 mini

SL · Oct 3, 2026

Source label: o3 mini (High)

verified runtimeexact aliasBackground only

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 9%
Last updated: recent
Eligibility: historical_model
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: o3 mini (High). Collapse policy: highest reported score per canonical model.

46.8%

#63 · kimi-k2-instruct

SL · Oct 3, 2026

verified runtimeexact directBackground only

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 7.5%
Last updated: recent
Eligibility: benchmark_derived_model
Identity: exact (1.00)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: kimi-k2-instruct. Collapse policy: highest reported score per canonical model.

46.7%

#64 · deepseek-v3.1

SL · Oct 3, 2026

Source label: DeepSeek-V3.1

verified runtimeexact alias

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 6%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: DeepSeek-V3.1. Collapse policy: highest reported score per canonical model.

46.3%

#65 · Deepseek V3 (March 2025)

SL · Oct 3, 2026

Source label: Deepseek V3 (March 2025)

verified runtimeexact aliasBackground only

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 4.5%
Last updated: recent
Eligibility: preview_model
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: Deepseek V3 (March 2025). Collapse policy: highest reported score per canonical model.

44.5%

#66 · Gemini 3 Pro Preview

SL · Oct 3, 2026

verified runtimeexact directBackground only

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 3%
Last updated: recent
Eligibility: preview_model
Identity: exact (1.00)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: gemini-3-pro-preview. Collapse policy: highest reported score per canonical model.

42.6%

#67 · Mistral Medium 3

SL · Oct 3, 2026

Source label: Mistral Medium 3

verified runtimeexact aliasBackground only

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 3%
Last updated: recent
Eligibility: historical_model
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: Mistral Medium 3. Collapse policy: highest reported score per canonical model.

42.6%

#68 · Gemini 3.1 Pro Preview

SL · Oct 3, 2026

Source label: gemini-3.1-pro-preview

verified runtimeexact aliasBackground only

Raw row drilldownsource, percentile, eligibility

Source URL: https://labs.scale.com/leaderboard/mask
Percentile: 0%
Last updated: recent
Eligibility: preview_model
Identity: provider alias (0.92)

Parsed from the public Scale Labs page for scale-mask. Reported model configuration: gemini-3.1-pro-preview. Collapse policy: highest reported score per canonical model.

42.4%

MASK

Test details

What it measures vs what it misses

✓ Measures

✗ Misses

Leaderboard · this benchmark version

Loading benchmark evidence.

MASK

Test details

What it measures vs what it misses

✓ Measures

✗ Misses

Leaderboard · this benchmark version