Skip to main content
BenchLM

Scale Labs MASK (MASK)

We show this table for reference; we do not rank on it.

A Scale Labs public leaderboard mirrored as display-only reference data. It does not affect BenchLM rankings.

Scale score on MASK — September 29, 2026 snapshot

We mirror the published scale score view for MASK. claude-opus-4-6 (Non-Thinking) leads the public snapshot at 96.3%, followed by Claude Sonnet 4.5 Thinking (96.1%) and Claude Sonnet 4 (Thinking) (95.3%). We do not use these results to rank models overall.

1

claude-opus-4-6 (Non-Thinking)

Anthropic

96.3%
2Proprietary

Claude Sonnet 4.5 Thinking

Anthropic

96.1%
Context 200K
3

Claude Sonnet 4 (Thinking)

Anthropic

95.3%

67 modelsInstruction FollowingCurrentDisplay onlyUpdated September 29, 2026 snapshot

Scale score table (67 models)

Score
1
claude-opus-4-6 (Non-Thinking)Anthropic
96.3%
2
Claude Sonnet 4.5 ThinkingAnthropic · Closed
96.1%
3
Claude Sonnet 4 (Thinking)Anthropic
95.3%
4
claude-opus-4-1-20250805-thinkingAnthropic
94.2%
5
Claude Opus 4.5 ThinkingAnthropic · Closed
92.5%
6
GPT-OSS 120BOpenAI · Open weight
92%
7
GPT-5.4 ProOpenAI · Closed
91.7%
8
gpt-5.4-2026-03-05 (xhigh thinking)OpenAI
89.7%
9
Claude Sonnet 4Anthropic
89.3%
10
Claude Opus 4 (Thinking)Anthropic
87.9%
11
claude-opus-4-1-20250805Anthropic
87.4%
12
Claude Opus 4.5Anthropic · Closed
87.1%
13
GPT-5.2OpenAI · Closed
86.7%
14
GPT-OSS 20BOpenAI · Open weight
86.5%
15
Claude Sonnet 4.5Anthropic · Closed
86.4%
16
gpt-5.1-thinkingOpenAI
86.3%
17
gpt-5-pro-2025-10-06OpenAI
86.0%
18
claude-opus-4-6-thinking-maxAnthropic
85.4%
19
o3 (high) (April 2025)OpenAI
84.5%
20
o3 (medium) (April 2025)OpenAI
82.6%
21
GPT-5 miniOpenAI · Closed
82.6%
22
o3 Pro (high) (June 2025)OpenAI
82.5%
23
Claude 3.7 Sonnet (Thinking) (February 2025)Anthropic
82.1%
24
Claude Opus 4Anthropic
80.3%
25
gpt-5-2025-08-07OpenAI
79.3%
26
Claude 3 OpusAnthropic · Closed
79%
27
o4-mini (high) (April 2025)OpenAI
78.6%
28
o4-mini (medium) (April 2025)OpenAI
72.9%
29
Claude 3.5 Sonnet (October 2024)Anthropic
72.3%
30
Claude 3.7 Sonnet (February 2025)Anthropic
72.3%
31
Kimi K2.5Moonshot AI · Open weight
70.5%
32
gpt-5.1-instantOpenAI
63.3%
33
o1-proOpenAI · Closed
61.6%
34
glm-4p5Z.AI
61.5%
35
GPT-4.1 nano\nOpenAI
61.4%
36
Llama 3.1 405B InstructMeta
61.4%
37
glm-4p5-airZ.AI
60.8%
38
gpt 4o (November 2024)OpenAI
60.1%
39
o1 (December 2024)OpenAI
59.3%
40
Deepseek R1 (Jan 2025)Deepseek
57.3%
41
GPT 4.5 PreviewOpenAI
56.9%
42
Mistral MagistralMistral
56.5%
43
Qwen3-235B-A22BAlibaba
56.4%
44
Gemini 2.5 Pro Experimental (March 2025)Google
55.9%
45
gemini-2.5-pro-preview-06-05Google
55.7%
46
Llama 3.2 90B Vision InstructMeta
54.1%
47
Gemini 2.5 Pro Preview (May 06 2025)Google
53.1%
48
DeepSeek-R1-0528Deepseek
53%
49
Llama 3.3 70B InstructMeta
51.9%
50
GPT-4.1OpenAI · Closed
51.1%
51
GPT-4.1 miniOpenAI · Closed
50%
52
Llama 4 MaverickMeta · Open weight
49.7%
53
o3 mini (Low)OpenAI
49.7%
54
Gemini 2.0 Flash Thinking (January 2025)Google
49.5%
55
Gemini 2.5 Flash Preview (May 2025)Google
49.1%
56
Gemini 2.0 FlashGoogle
49.1%
57
o3 mini (Medium)OpenAI
48.9%
58
Gemini 2.0 Pro Experimental (February 2025)Google
48.7%
59
gemini-3.1-flash-lite-previewGoogle
48.4%
60
Mistral Large 2411Mistral
47.5%
61
o3 mini (High)OpenAI
46.8%
62
kimi-k2-instructMoonshot AI
46.7%
63
DeepSeek V3.1DeepSeek · Open weight
46.3%
64
Deepseek V3 (March 2025)Deepseek
44.5%
65
gemini-3-pro-previewGoogle
42.6%
66
Mistral Medium 3Mistral · Closed
42.6%
67
gemini-3.1-pro-previewGoogle
42.4%

How to read this leaderboard

Compare the published configurations as complete evaluation systems. The source can combine a base model, agent scaffold, tools, budget, and inference setting in each result.

Operator receipt: 67 sourced rows are currently displayable on this page; the leading published row is claude-opus-4-6 (Non-Thinking) at 96.3%.

Honest limit: This Scale table is display-only context, not benchmark provenance or a weighted model-only comparison.

MASK on Scale Labs

How BenchLM shows MASK

BenchLM mirrors 67 published rows from Scale Labs’ public MASK leaderboard, captured on September 29, 2026 snapshot.

The table is display only. It is useful context for a published agent or model configuration, but it does not enter BenchLM’s overall or category rankings.

Snapshot

67 published rowsScale Labs sourceDisplay only
MASK on Scale Labs

The published MASK snapshot places claude-opus-4-6 (Non-Thinking) first at 96.3%. The third row is 1.0 points behind. The broader top-10 range is 8.4 points, so many of the published results sit in a relatively narrow band.

67 models have been evaluated on MASK. The benchmark falls in the Instruction Following category. We keep external benchmark mirrors separate from the weighted global scoring system, so these results remain source-specific evidence. MASK is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.

About MASK

Year

2026

Tasks

67 published rows

Format

Published Scale leaderboard score

Difficulty

External agent and model evaluation

BenchLM mirrors 67 published rows from the MASK public table captured on September 29, 2026 snapshot.

Scale Labs leaderboardPublic benchmark source

Freshness and provenance

Version

MASK 2026

Refresh cadence

Quarterly

Staleness state

Current

Question availability

Public benchmark set

CurrentDisplay only

BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.

Questions

What does MASK measure?

A Scale Labs public leaderboard mirrored as display-only reference data. It does not affect BenchLM rankings.

Which model leads the published MASK snapshot?

claude-opus-4-6 (Non-Thinking) currently leads the published MASK snapshot with 96.3% scale score. BenchLM shows this benchmark for display only and does not use it in overall rankings.

How many models are evaluated on MASK?

The September 29, 2026 snapshot snapshot contains 67 AI models.

Last updated: September 29, 2026 snapshot · mirrored from the public benchmark leaderboard

Know when it’s worth switching models

The model to choose, the cheaper alternative, and the release we would wait on.

Read a sample issue

Join 2,000+ readers.

One email each week. Unsubscribe anytime.