AnthropicReasoning model

Claude 3.7 Sonnet

Name: Claude 3.7 Sonnet benchmark results
Creator: BenchLeader

Best configuration ranks #255 of 610 on the BenchLeader Index at 52.1 ±3.0 (thinking reasoning effort). Last measured 2 Sept 2026. Released 24 Feb 2025.

Blended price: $6.00/M; $3.00 in · $15.00 out
Output speed: –
First answer: –
Context: 200k

How it scores by categoryDashed line = average model (50). One step of 15 = one standard deviation.

Overall index52
Reasoning58
Coding47
Agents & tools52
Maths49
Knowledge55
Instruction following46
Human preference57
Multimodal55
Composite51

Reasoning-effort configurations

The same model behaves differently depending on how much it is allowed to think. Each row is one setting, scored only on the benchmarks that were run at that setting. “Default” means the publisher did not say which setting was used.

Effort	Index	Rank	Speed	First answer	Chat reply cost	Categories
no reasoning	–	–	–	–	$0.0057	Coding 59
thinkingbest	52.1	#255	–	–	$0.0057	Agents & tools 52 · Coding 47 · Composite 51 · Human preference 57 · Instruction following 46 · Knowledge 55 · Maths 49 · Multimodal 55 · Reasoning 58
default	49.8	#300	–	–	$0.0057	Agents & tools 48 · Coding 47 · Composite 48 · Human preference 55 · Instruction following 46 · Knowledge 54 · Long context 50 · Maths 51 · Multimodal 50 · Reasoning 49

Benchmark results

One column per reasoning effort. Rank is among every configuration of every model on that benchmark. Hover a score for the run it came from.

Reasoning

Benchmark	no reasoning	thinking	default	Source
GPQA Diamond	–	–	79.7%#115	Epoch AI Benchmarking Hub
Humanity's Last Exam	–	–	8.0%#31	Scale AI / CAIS
SimpleBench	–	–	46.4%#51	SimpleBench
LMArena Hard Prompts	–	1417#135	1398#154	LMArena
GPQA Diamond (AA)not in index	–	77.2%#215	65.6%#326	Artificial Analysis
Humanity's Last Exam (AA)not in index	–	9.7%#264	4.2%#452	Artificial Analysis
GPQA Diamond (Vals)not in index	–	75.3%#78	67.4%#97	Vals AI

Coding

Benchmark	no reasoning	thinking	default	Source
SWE-bench Verified (Epoch)	–	–	61.0%#29	Epoch AI Benchmarking Hub
GSO-Bench	–	–	3.8%#26	GSO-Bench
LMArena Coding	–	1452#125	1431#151	LMArena
LiveCodeBench	–	60.4%#102	56.7%#107	Vals AI
Aider Polyglot	60.4%#14	–	64.9%#10	Aider polyglot leaderboard
SWE-bench Verified (bash only)	–	–	52.8%#32	SWE-bench
SWE-bench Verified (any scaffold)not in index	–	–	70.4%#16	SWE-bench

Agents & tools

Benchmark	no reasoning	thinking	default	Source
Cybench	–	–	20.0%#11	Cybench
Terminal-Bench Hard	–	21.2%#152	21.2%#152	Artificial Analysis
τ²-Bench Telecom (AA)not in index	–	54.7%#180	50.0%#191	Artificial Analysis

Maths

Benchmark	no reasoning	thinking	default	Source
OTIS Mock AIME	–	–	57.8%#150	Epoch AI Benchmarking Hub
MATH Level 5	–	–	91.2%#18	Epoch AI Benchmarking Hub
AIME (Vals)	–	44.6%#62	22.3%#77	Vals AI
MGSM	–	93.0%#12	92.4%#19	Vals AI

Knowledge

Benchmark	no reasoning	thinking	default	Source
AA-Omniscience	–	-0.7#99	-9.7#136	Artificial Analysis
MMLU-Pro	–	82.7%#74	80.7%#82	Vals AI
LegalBench	–	82.5%#61	80.0%#83	Vals AI
CorpFin	–	60.4%#68	–	Vals AI
TaxEval	–	74.0%#41	72.4%#62	Vals AI
MedQA	–	90.2%#46	–	Vals AI
MultiNRC	–	27.8%#26	–	Scale AI SEAL

Instruction following

Benchmark	no reasoning	thinking	default	Source
IFBench	–	48.3%#181	44.0%#214	Artificial Analysis
MultiChallenge	–	51.6%#21	–	Scale AI SEAL

Human preference

Benchmark	no reasoning	thinking	default	Source
LMArena Text	–	1388#144	1372#160	LMArena

Multimodal

Benchmark	no reasoning	thinking	default	Source
LMArena Vision	–	1169#89	1151#98	LMArena
MMMU-Pro	–	–	60.1%#178	Artificial Analysis
VISTA	–	48.2%#16	43.0%#35	Scale AI SEAL
MMMU (validation)	–	–	75.0%#14	MMMU

Long context

Benchmark	no reasoning	thinking	default	Source
Fiction.LiveBench 120k	–	–	53.1%#20	Fiction.live
AA-LCR	–	–	51.7%#252	Artificial Analysis

Composite

Benchmark	no reasoning	thinking	default	Source
Epoch Capabilities Indexnot in index	–	–	141.2#94	Epoch AI Benchmarking Hub
AA Intelligence Index	–	17.7#197	15.3#219	Artificial Analysis

What a task costs

Estimates from list price, output speed and time to first answer for the best configuration. “With caching” assumes three-quarters of the input is served from the prompt cache. Reasoning tokens are not modelled.

Workload	Tokens in / out	Cost	With caching	Time
Chat reply	400 / 300	$0.0057	–	–
Summarise a 30-page report	12,000 / 600	$0.045	–	–
Code edit	6,000 / 1,500	$0.041	–	–
Agentic coding session	60,000 / 4,000	$0.240	–	–
Structured extraction	2,000 / 200	$0.0090	–	–