Anthropicmodel systemcard
Claude Sonnet 5
#7of 27 · general
adaptive-reasoning efforts collapsed
Standing
dot = this model · shaded = the field's middle 60% · track = full fieldGeneral
60.3#7/27
Math
61.7#8/27
Coding
62.8#7/27
Health
62.0#8/26
Engineering
62.5#7/17
Analyst
60.0#5/11
Legal
36.8#19/26
Instruct
36.1#15/27
Agents
82.8#2/27
Professional
58.6#8/27
Multimodal
41.8#10/19
Hallucination
23.0#18/27
Speed
56.9#9/22
Value
49.9#12/27
Measured
≈ imputed · amber dot = lab-self-reported among the sourcesGPQA Diamond90.5%#15
HLE49.3%#6
MMLU-Pro87.5%#12
LiveBench76.0%#8
SWE-bench V82.4%#7
LiveCodeBench82.4%#17
SciCode53.6%#10
CritPt16.9%#10
Terminal-Bench80.4%#6
APEX Agents32.5%#12
GDPval54.9%#7
Arena1460#15
AA Index55.3#9
Vals Index68.6#6
Epoch Capabilities Index155.5#9
Operations
$2 / $10$ per 1M tokens, in / out
$4blended
73tokens / sec
$0.51$ per solved task
1460arena elo · 15,622 votes