Anthropicmodel systemcard

Claude Opus 5

#2of 27 · general

max/high effort collapsed

1Mcontext

Proprietarynon-commercial

2026-07-24released

paper ↗

Standing

dot = this model · shaded = the field's middle 60% · track = full field

General

89.2#2/27

Math

86.8#2/27

Coding

78.7#3/27

Health

86.2#2/26

Engineering

100.0#1/17

Cyber

7.7#13/14

Analyst

100.0#1/11

Legal

67.4#4/26

Instruct

51.5#11/27

Agents

97.4#1/27

Professional

100.0#1/27

Multimodal

91.7#1/19

Hallucination

55.7#3/27

Speed

30.5#17/22

Value

68.7#3/27

Measured

≈ imputed · amber dot = lab-self-reported among the sources

GPQA Diamond93.7%#3

HLE59.8%#2

MMLU-Pro91.6%#1

LiveBench80.1%#4

SWE-bench V97.0%#1

LiveCodeBench89.0%#4

SciCode55.7%#7

CritPt29.1%#3

Terminal-Bench86.9%#2

APEX Agents43.5%#2

GDPval67.4%#1

Arena1495#2

AA Index63.0#1

Vals Index74.8#2

Epoch Capabilities Index161.0#3

Operations

$5 / $25$ per 1M tokens, in / out

$10blended

53tokens / sec

$0.70$ per solved task

1495arena elo · 2,386 votes

Series

Anthropic's models on the board, by release

Claude Fable 52026-06-09 · 96Claude Sonnet 52026-06-30 · 60Claude Opus 52026-07-24 · 89

Neighbors

nearest capability fingerprints · deltas describe the neighbor

Claude Fable 5→ follows instructions better · pricier Muse Spark→ weaker math · cheaper GPT-5.6 Sol→ weaker agent Kimi K3→ cheaper GLM-5.2→ cheaper