AI Model Benchmark Comparison

Back to timeline

General Knowledge% Benchmark Score4/4 reported

Artificial Analysis Intelligence Index

LeaderClaude Opus 5.557.6%

+1.6% over Claude Sonnet 5.5

1Claude Opus 5.5

57.6%Indep.

2Claude Sonnet 5.5

56.0%Indep.

3Kimi K3.1 Pro

55.1%Indep.

4Claude Fable 5.1

53.4%Indep.

Coding% Benchmark Score1/4 reported

Artificial Analysis Coding Index

LeaderClaude Fable 5.181.6%

Sole evaluated model in selection

1Claude Fable 5.1

81.6%Indep.

2Kimi K3.1 Pro

Not reported

—

3Claude Sonnet 5.5

Not reported

—

4Claude Opus 5.5

Not reported

—

Agentic% Benchmark Score1/4 reported

Artificial Analysis Agentic Index

LeaderClaude Fable 5.157.9%

Sole evaluated model in selection

1Claude Fable 5.1

57.9%Indep.

2Kimi K3.1 Pro

Not reported

—

3Claude Sonnet 5.5

Not reported

—

4Claude Opus 5.5

Not reported

—

Coding% Benchmark Score3/4 reported

Artificial Analysis Coding Agent Index v2.0

LeaderClaude Opus 5.561.7%

+2.4% over Claude Sonnet 5.5

1Claude Opus 5.5

61.7%Indep.

2Claude Sonnet 5.5

59.3%Indep.

3Claude Fable 5.1

58.7%Indep.

4Kimi K3.1 Pro

Not reported

—

Coding% Benchmark Score3/4 reported

SWE-bench Pro

LeaderClaude Opus 5.589.9%

+8.6% over Claude Sonnet 5.5

1Claude Opus 5.5

89.9%Official

2Claude Sonnet 5.5

81.3%Official

3Claude Fable 5.1

81.2%Official

4Kimi K3.1 Pro

Not reported

—

Agentic% Benchmark Score4/4 reported

Terminal-Bench 4.0

LeaderClaude Sonnet 5.570.6%

+4.2% over Claude Opus 5.5

1Claude Sonnet 5.5

70.6%Official

2Claude Opus 5.5

66.4%Official

3Kimi K3.1 Pro

65.7%Official

4Claude Fable 5.1

55.8%Official

Reasoning% Benchmark Score3/4 reported

ARC-AGI-2

LeaderClaude Opus 5.591.7%

+1.7% over Claude Fable 5.1

1Claude Opus 5.5

91.7%Indep.

2Claude Fable 5.1

90.0%Official

3Claude Sonnet 5.5

83.9%Official

4Kimi K3.1 Pro

Not reported

—

Reasoning% Benchmark Score1/4 reported

GPQA Diamond

LeaderKimi K3.1 Pro94.2%

Sole evaluated model in selection

1Kimi K3.1 Pro

94.2%Official

2Claude Sonnet 5.5

Not reported

—

3Claude Opus 5.5

Not reported

—

4Claude Fable 5.1

Not reported

—

Reasoning% Benchmark Score4/4 reported

Humanity's Last Exam

LeaderClaude Opus 5.564.4%

+3.5% over Claude Fable 5.1

1Claude Opus 5.5

64.4%Official

2Claude Fable 5.1

60.9%Official

3Claude Sonnet 5.5

56.9%Official

4Kimi K3.1 Pro

51.1%Official

Mathematics% Benchmark Score3/4 reported

FrontierMath v2 (Tiers 1-3)

LeaderClaude Opus 5.591.2%

+1.0% over Claude Fable 5.1

1Claude Opus 5.5

91.2%Indep.

2Claude Fable 5.1

90.2%Indep.

3Claude Sonnet 5.5

88.8%Indep.

4Kimi K3.1 Pro

Not reported

—

Human PreferenceELO Rating4/4 reported

Design Arena ELO

LeaderClaude Opus 5.51468

+20 ELO over Kimi K3.1 Pro

1Claude Opus 5.5

1468Indep.

2Kimi K3.1 Pro

1448Indep.

3Claude Fable 5.1

1416Indep.

4Claude Sonnet 5.5

1413Indep.

Responsible Benchmark Interpretation

Evaluation Methodology & Contamination Safeguards

Workload Specialization

Leading on coding evaluations (SWE-bench Pro, Terminal-Bench) does not imply superior creative writing or conversational nuance. Choose models tailored to your specific application workload.

Latency vs Compute Costs

Frontier reasoning models consume extended test-time compute. Compare the Artificial Analysis Intelligence Index against output price to find the best ROI for high-volume or interactive workloads.

Contamination & Satiation

Legacy benchmarks (MATH Level 5, GSM8K, HumanEval, MMLU-Pro) have saturated or stopped being reported. This catalogue tracks the boards labs still publish: SWE-bench Pro, Terminal-Bench 4.0, FrontierMath v2, ARC-AGI-2, GPQA Diamond and HLE.

Coverage ledger

What is actually reported?

87/137 models · 262 source URLs

Artificial Analysis Intelligence Index72/137

Artificial Analysis Coding Index50/137

Design Arena ELO46/137

Artificial Analysis Agentic Index45/137

FrontierMath v2 (Tiers 1-3)41/137

SWE-bench Pro36/137

Terminal-Bench 4.035/137

ARC-AGI-234/137

GPQA Diamond33/137

Humanity's Last Exam32/137

Artificial Analysis Coding Agent Index v2.018/137

442 source-backed scores across 87 models. 50 models have no published result in the tracked boards yet; they are not scored by inference.

Catalogue audited 27 Oct 2026

Verified JSON Telemetry

Loading benchmark catalogue

Back to timeline

General Knowledge% Benchmark Score4/4 reported

Artificial Analysis Intelligence Index

LeaderClaude Opus 5.557.6%

+1.6% over Claude Sonnet 5.5

1Claude Opus 5.5

57.6%Indep.

2Claude Sonnet 5.5

56.0%Indep.

3Kimi K3.1 Pro

55.1%Indep.

4Claude Fable 5.1

53.4%Indep.

Coding% Benchmark Score1/4 reported

Artificial Analysis Coding Index

LeaderClaude Fable 5.181.6%

Sole evaluated model in selection

1Claude Fable 5.1

81.6%Indep.

2Kimi K3.1 Pro

Not reported

—

3Claude Sonnet 5.5

Not reported

—

4Claude Opus 5.5

Not reported

—

Agentic% Benchmark Score1/4 reported

Artificial Analysis Agentic Index

LeaderClaude Fable 5.157.9%

Sole evaluated model in selection

1Claude Fable 5.1

57.9%Indep.

2Kimi K3.1 Pro

Not reported

—

3Claude Sonnet 5.5

Not reported

—

4Claude Opus 5.5

Not reported

—

Coding% Benchmark Score3/4 reported

Artificial Analysis Coding Agent Index v2.0

LeaderClaude Opus 5.561.7%

+2.4% over Claude Sonnet 5.5

1Claude Opus 5.5

61.7%Indep.

2Claude Sonnet 5.5

59.3%Indep.

3Claude Fable 5.1

58.7%Indep.

4Kimi K3.1 Pro

Not reported

—

Coding% Benchmark Score3/4 reported

SWE-bench Pro

LeaderClaude Opus 5.589.9%

+8.6% over Claude Sonnet 5.5

1Claude Opus 5.5

89.9%Official

2Claude Sonnet 5.5

81.3%Official

3Claude Fable 5.1

81.2%Official

4Kimi K3.1 Pro

Not reported

—

Agentic% Benchmark Score4/4 reported

Terminal-Bench 4.0

LeaderClaude Sonnet 5.570.6%

+4.2% over Claude Opus 5.5

1Claude Sonnet 5.5

70.6%Official

2Claude Opus 5.5

66.4%Official

3Kimi K3.1 Pro

65.7%Official

4Claude Fable 5.1

55.8%Official

Reasoning% Benchmark Score3/4 reported

ARC-AGI-2

LeaderClaude Opus 5.591.7%

+1.7% over Claude Fable 5.1

1Claude Opus 5.5

91.7%Indep.

2Claude Fable 5.1

90.0%Official

3Claude Sonnet 5.5

83.9%Official

4Kimi K3.1 Pro

Not reported

—

Reasoning% Benchmark Score1/4 reported

GPQA Diamond

LeaderKimi K3.1 Pro94.2%

Sole evaluated model in selection

1Kimi K3.1 Pro

94.2%Official

2Claude Sonnet 5.5

Not reported

—

3Claude Opus 5.5

Not reported

—

4Claude Fable 5.1

Not reported

—

Reasoning% Benchmark Score4/4 reported

Humanity's Last Exam

LeaderClaude Opus 5.564.4%

+3.5% over Claude Fable 5.1

1Claude Opus 5.5

64.4%Official

2Claude Fable 5.1

60.9%Official

3Claude Sonnet 5.5

56.9%Official

4Kimi K3.1 Pro

51.1%Official

Mathematics% Benchmark Score3/4 reported

FrontierMath v2 (Tiers 1-3)

LeaderClaude Opus 5.591.2%

+1.0% over Claude Fable 5.1

1Claude Opus 5.5

91.2%Indep.

2Claude Fable 5.1

90.2%Indep.

3Claude Sonnet 5.5

88.8%Indep.

4Kimi K3.1 Pro

Not reported

—

Human PreferenceELO Rating4/4 reported

Design Arena ELO

LeaderClaude Opus 5.51468

+20 ELO over Kimi K3.1 Pro

1Claude Opus 5.5

1468Indep.

2Kimi K3.1 Pro

1448Indep.

3Claude Fable 5.1

1416Indep.

4Claude Sonnet 5.5

1413Indep.

Responsible Benchmark Interpretation

Evaluation Methodology & Contamination Safeguards

Workload Specialization

Leading on coding evaluations (SWE-bench Pro, Terminal-Bench) does not imply superior creative writing or conversational nuance. Choose models tailored to your specific application workload.

Latency vs Compute Costs

Frontier reasoning models consume extended test-time compute. Compare the Artificial Analysis Intelligence Index against output price to find the best ROI for high-volume or interactive workloads.

Contamination & Satiation

Coverage ledger

What is actually reported?

87/137 models · 262 source URLs

Artificial Analysis Intelligence Index72/137

Artificial Analysis Coding Index50/137

Design Arena ELO46/137

Artificial Analysis Agentic Index45/137

FrontierMath v2 (Tiers 1-3)41/137

SWE-bench Pro36/137

Terminal-Bench 4.035/137

ARC-AGI-234/137

GPQA Diamond33/137

Humanity's Last Exam32/137

Artificial Analysis Coding Agent Index v2.018/137

442 source-backed scores across 87 models. 50 models have no published result in the tracked boards yet; they are not scored by inference.

Catalogue audited 27 Oct 2026 (2d)

Verified JSON Telemetry

General Knowledge% Benchmark Score4/4 reported

Artificial Analysis Intelligence Index

LeaderClaude Opus 5.557.6%

+1.6% over Claude Sonnet 5.5

1Claude Opus 5.5

57.6%Indep.

2Claude Sonnet 5.5

56.0%Indep.

3Kimi K3.1 Pro

55.1%Indep.

4Claude Fable 5.1

53.4%Indep.

Coding% Benchmark Score1/4 reported

Artificial Analysis Coding Index

LeaderClaude Fable 5.181.6%

Sole evaluated model in selection

1Claude Fable 5.1

81.6%Indep.

2Kimi K3.1 Pro

Not reported

—

3Claude Sonnet 5.5

Not reported

—

4Claude Opus 5.5

Not reported

—

Agentic% Benchmark Score1/4 reported

Artificial Analysis Agentic Index

LeaderClaude Fable 5.157.9%

Sole evaluated model in selection

1Claude Fable 5.1

57.9%Indep.

2Kimi K3.1 Pro

Not reported

—

3Claude Sonnet 5.5

Not reported

—

4Claude Opus 5.5

Not reported

—

Coding% Benchmark Score3/4 reported

Artificial Analysis Coding Agent Index v2.0

LeaderClaude Opus 5.561.7%

+2.4% over Claude Sonnet 5.5

1Claude Opus 5.5

61.7%Indep.

2Claude Sonnet 5.5

59.3%Indep.

3Claude Fable 5.1

58.7%Indep.

4Kimi K3.1 Pro

Not reported

—

Coding% Benchmark Score3/4 reported

SWE-bench Pro

LeaderClaude Opus 5.589.9%

+8.6% over Claude Sonnet 5.5

1Claude Opus 5.5

89.9%Official

2Claude Sonnet 5.5

81.3%Official

3Claude Fable 5.1

81.2%Official

4Kimi K3.1 Pro

Not reported

—

Agentic% Benchmark Score4/4 reported

Terminal-Bench 4.0

LeaderClaude Sonnet 5.570.6%

+4.2% over Claude Opus 5.5

1Claude Sonnet 5.5

70.6%Official

2Claude Opus 5.5

66.4%Official

3Kimi K3.1 Pro

65.7%Official

4Claude Fable 5.1

55.8%Official

Reasoning% Benchmark Score3/4 reported

ARC-AGI-2

LeaderClaude Opus 5.591.7%

+1.7% over Claude Fable 5.1

1Claude Opus 5.5

91.7%Indep.

2Claude Fable 5.1

90.0%Official

3Claude Sonnet 5.5

83.9%Official

4Kimi K3.1 Pro

Not reported

—

Reasoning% Benchmark Score1/4 reported

GPQA Diamond

LeaderKimi K3.1 Pro94.2%

Sole evaluated model in selection

1Kimi K3.1 Pro

94.2%Official

2Claude Sonnet 5.5

Not reported

—

3Claude Opus 5.5

Not reported

—

4Claude Fable 5.1

Not reported

—

Reasoning% Benchmark Score4/4 reported

Humanity's Last Exam

LeaderClaude Opus 5.564.4%

+3.5% over Claude Fable 5.1

1Claude Opus 5.5

64.4%Official

2Claude Fable 5.1

60.9%Official

3Claude Sonnet 5.5

56.9%Official

4Kimi K3.1 Pro

51.1%Official

Mathematics% Benchmark Score3/4 reported

FrontierMath v2 (Tiers 1-3)

LeaderClaude Opus 5.591.2%

+1.0% over Claude Fable 5.1

1Claude Opus 5.5

91.2%Indep.

2Claude Fable 5.1

90.2%Indep.

3Claude Sonnet 5.5

88.8%Indep.

4Kimi K3.1 Pro

Not reported

—

Human PreferenceELO Rating4/4 reported

Design Arena ELO

LeaderClaude Opus 5.51468

+20 ELO over Kimi K3.1 Pro

1Claude Opus 5.5

1468Indep.

2Kimi K3.1 Pro

1448Indep.

3Claude Fable 5.1

1416Indep.

4Claude Sonnet 5.5

1413Indep.

Responsible Benchmark Interpretation

Evaluation Methodology & Contamination Safeguards

Workload Specialization

Leading on coding evaluations (SWE-bench Pro, Terminal-Bench) does not imply superior creative writing or conversational nuance. Choose models tailored to your specific application workload.

Latency vs Compute Costs

Frontier reasoning models consume extended test-time compute. Compare the Artificial Analysis Intelligence Index against output price to find the best ROI for high-volume or interactive workloads.

Contamination & Satiation

Coverage ledger

What is actually reported?

87/137 models · 262 source URLs

Artificial Analysis Intelligence Index72/137

Artificial Analysis Coding Index50/137

Design Arena ELO46/137

Artificial Analysis Agentic Index45/137

FrontierMath v2 (Tiers 1-3)41/137

SWE-bench Pro36/137

Terminal-Bench 4.035/137

ARC-AGI-234/137

GPQA Diamond33/137

Humanity's Last Exam32/137

Artificial Analysis Coding Agent Index v2.018/137

442 source-backed scores across 87 models. 50 models have no published result in the tracked boards yet; they are not scored by inference.

Catalogue audited 27 Oct 2026

Verified JSON Telemetry