LLM Boss — Frontier LLM Benchmarks Compared

Fable 5.1vsGPT-6 Astra

Fable 5.1

$10/$50 per Mtok

Opus 5.5

$4/$20 per Mtok

Sonnet 5.5

$2/$10 per Mtok

GPT-6 Astra

$10/$50 per Mtok

GPT-6.1 Sol

$2/$10 per Mtok

Grok 4.7

$2/$6 per Mtok

DeepSWE

—

74.2%

71.0%

74.1%

75.2%

71.0%

T-Bench 4.0

57.9%

64.8%

61.8%

58.2%

58.2%

37.6%

AA-LCR v1.1

85.3%

84.7%

82.7%

80.7%

83.0%

76.7%

Finance Agent v2

58.9%

58.6%

58.1%

53.5%

52.0%

52.3%

GDPval-AA v2.1

1756

1866

1839

1542

1575

1712

HealthBench Prof.

62.1%

65.6%

69.2%

64.7%

64.2%

56.7%

SWE-bench Pro

81.2%

89.9%

81.3%

—

—

—

SWE-bench ML

89.1%

93.9%

90.3%

—

—

—

FrontierCode 1.1

50.3%

54.4%

46.2%

53.3%

—

—

SWE-bench Verified

—

—

—

—

—

—

HLE (tools)

65.6%

67.7%

64.5%

57.2%

—

—

BrowseComp

—

—

—

91.5%

—

—

MCP-Atlas

—

—

—

—

—

—

AutomationBench

—

42.5%

44.7%

41.4%

36.1%

—

OSWorld 2.1

80.7%

partial

42.8%

strict

81.8%

partial

48.7%

strict

80.1%

partial

43.5%

strict

—

—

—

Blueprint-Bench 2

—

—

—

—

—

—

GDPpdf

—

—

—

—

—

—

Legal Agent

—

8.3%

10.0%

—

—

—

CyberGym

—

—

—

—

—

—

ExploitBench

—

91.0%

80.0%

—

—

—

BioMysteryBench

—

89.3%

human solvable

50.0%

human difficult

—

—

—

—

GPQA Diamond

—

—

—

96.0%

—

—

Chartography

44.8%

no tools

88.4%

with tools

64.4%

no tools

89.0%

with tools

61.6%

no tools

90.2%

with tools

—

—

—

MMMLU

—

—

—

—

—

—

Showing 6 of 24 benchmarks. A dash means the lab has not published a comparable figure.