Scale Labs

[LEADERBOARDS]

Testing the limits of AI.

Benchmarks for frontier, agentic, and safety capabilities

Benchmarks20+

Including benchmarks on agentic coding, frontier reasoning, and safety alignment.

Models evaluated100+

From leading AI labs including OpenAI, Anthropic, Google, Meta, and open-source contributors.

Humanity's Sixth Sense

HSS, in partnership with Elorian, is a benchmark for intuitive visual reasoning across diverse image and video inputs.

1

Gemini 4 Carbon (high)

NEW

71.10±3.60

2

Claude Fable 5.2 (max)

NEW

59.50±3.80

3

Gemini-4-Argon (high)

NEW

55.40±4.20

View Full Ranking →

DistressBench

We introduce DistressBench, designed to measure the quality of crisis support a model delivers.

1

muse-spark-1.2

88.39±0.93

2

muse-spark-open

NEW

87.74±1.03

3

muse-wonder-1.0

NEW

85.89±1.05

View Full Ranking →

Humanity's Last Exam (Diamond)

Challenging LLMs at the frontier of human knowledge

1

claude-fable-5-2

NEW

72.40±2.80

2

60.60±3.00

3

55.00±3.10

View Full Ranking →

DrugDiscoveryBench

DrugDiscoveryBench: 82 expert-curated tasks evaluating how reliably frontier coding agents perform the computational work of early-stage drug discovery.

1

Gemini 4 Carbon (mini-swe-agent) high

NEW

71.50±2.10

2

Gemini 4 Hydrogen (mini-swe-agent) high

NEW

69.10±3.50

3

GPT 6 Astra (mini-SWE-agent) max

NEW

68.70±2.50

View Full Ranking →

HiL-Bench (Human-in-Loop Benchmark)

Evaluates whether agents recognize information gaps and ask targeted clarifying questions.

1

Claude Fable 5.2 (*public set only)

NEW

80.00±4.95

2

Muse Wonder 1.0

79.67±4.17

3

GPT 6.1 Terra

68.00±5.02

View Full Ranking →

MCP Atlas

Evaluating real-world tool use through the Model Context Protocol (MCP)

1

Gemini-4-Argon

NEW

90.70±1.80

2

Muse Spark 1.1

88.10±1.95

3

Fable 5.1

87.20±2.05

View Full Ranking →

SWE-Bench Pro V2

Evaluating long-horizon software engineering tasks in public open source repositories

1

Opus 5 (Claude Code) xhigh

99.40±0.40

2

Gemini 4 Argon (mini-swe-agent) high

NEW

99.22±0.45

2

Fable 5.1 (Claude Code) high

99.10±0.50

1

Opus 5 (Claude Code) xhigh

98.00

2

Fable 5.1 (Claude Code) high

92.20

3

Gemini 4 Argon (mini-swe-agent) high

NEW

92.16

View Full Ranking →

SWE Atlas

SWE Atlas evaluates coding agents on codebase comprehension, test writing and refactoring

1

Fable 5.2

NEW

80.55±4.80

2

Opus 5.5

77.45±5.05

3

Fable 5.1

76.45±5.16

1

Fable 5.2

NEW

83.87±6.56

2

Qwen 3.8 Max

NEW

81.45±6.94

3

Fable 5.1

79.03±7.27

1

Fable 5.1

78.89±8.60

2

Gemini 4 Argon

77.78±8.76

3

Opus 5.5

77.78±8.76

1

Fable 5.2

NEW

80.00±9.61

2

Opus 5.5

77.14±10.08

3

GPT-6 Astra

71.43±10.85

View Full Ranking →

AudioMultiChallenge

Evaluating spoken dialogue systems in multi-turn interaction

1

gemini-3.8-flash (high)

NEW

60.40±4.49

2

salesforce-sov-omni-1

NEW

59.69±4.13

3

gemini-4-argon (high)

NEW

58.41±4.53

1

gpt-realtime-2 (xHigh)

48.45±4.59

2

tml-interaction-small

43.36±4.55

3

gpt-realtime-2

37.61±4.45

View Full Ranking →

Professional Reasoning Benchmark

Evaluating Professional Reasoning in Finance and Legal Practice

1

Muse Spark 1.3

NEW

59.54±1.66

2

Muse Wonder 1.0

NEW

58.67±1.69

3

Muse Spark 1.1

55.01±0.14

View Full Ranking →

Remote Labor Index (RLI)

Evaluating AI agents ability to perform real-world, economically valuable remote work

1

Fable 5.2

NEW

41.25

2

Opus 5.5

21.25

3

GPT 6 Astra

20.83

View Full Ranking →

MultiChallenge

Assessing models across diverse, interdisciplinary challenges

1

gpt 6.1 Sol

NEW

81.95±4.16

2

Gemini 4 Argon

NEW

78.70±5.30

3

Muse Spark

75.52±4.05

1

Nemotron 3.5 Bolt E4B (Thinking)

NEW

56.03±6.20

2

Gemma 4 E4B (Thinking)

43.23±6.13

3

Nemotron 3 Nano 4B (Thinking)

38.67±6.27

View Full Ranking →

Fortress

Frontier Risk Evaluation for National Security and Public Safety

1

5.78±1.26

2

7.64±1.15

3

8.24±1.93

View Full Ranking →

Frontier AI Model Evaluations & Benchmarks

We conduct high-complexity evaluations to expose model failures, prevent benchmark saturation, and push model capabilities--while continuously evaluating the latest frontier models.

Scaling with Human Expertise

Humans design complex evaluations and define precise criteria to assess models, while LLMs scale evaluations--ensuring efficiency and alignment with human judgment.

Robust Datasets for Reliable AI Benchmarks

Our leaderboards are built on carefully curated evaluation sets, combining private datasets to prevent overfitting and open-source datasets for broad benchmarking and comparability.

[EVALUATE YOUR MODEL]

If you'd like to add your model to this leaderboard or a future version, please contact [email protected]. To ensure leaderboard integrity, we require that models can only be featured the FIRST TIME when an organization encounters the prompts.