AnthropicReasoning model

Claude Sonnet 4.5

Name: Claude Sonnet 4.5 benchmark results
Creator: BenchLeader

Best configuration ranks #137 of 610 on the BenchLeader Index at 57.8 ±7.1 (high reasoning effort). Last measured 8 Sept 2026. Released 29 Sept 2025.

Blended price: $6.00/M; $3.00 in · $15.00 out
Output speed: 40 tok/s
First answer: 1.38 s; first token 1.38 s
Context: 200k

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

Overall index58
Reasoning66
Coding56
Human preference65

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
highbest	57.8	#137	40 tok/s	1.38 s	$0.0057	Coding 56 · Human preference 65 · Reasoning 66
thinking	52.8	#237	44 tok/s	17 s	$0.0057	Agents & tools 46 · Coding 50 · Composite 56 · Instruction following 51 · Knowledge 59 · Long context 63 · Maths 54 · Multimodal 57 · Reasoning 44
default	52.3	#249	40 tok/s	1.38 s	$0.0057	Agents & tools 58 · Coding 54 · Composite 53 · Human preference 65 · Instruction following 42 · Knowledge 49 · Long context 53 · Maths 47 · Multimodal 55 · 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	high	thinking	default	Source
GPQA Diamond	–	–	82.3%#101	Epoch AI Benchmarking Hub
Humanity's Last Exam	–	–	13.7%#23	Scale AI / CAIS
SimpleBench	–	–	54.3%#40	SimpleBench
LMArena Hard Prompts	1487#46	–	1483#50	LMArena
GPQA Diamond (AA)not in index	–	83.4%#157	72.7%#264	Artificial Analysis
Humanity's Last Exam (AA)not in index	–	17.8%#175	7.2%#300	Artificial Analysis
GPQA Diamond (Vals)not in index	–	81.6%#61	–	Vals AI
Kagi LLM Benchmark	–	–	57.9%#55	Kagi LLM Benchmark
ARC-AGI-1	–	63.7%#104	25.5%#150	ARC Prize
ARC-AGI-2	–	13.6%#99	3.8%#131	ARC Prize

Coding

Benchmark	high	thinking	default	Source
SWE-bench Verified (Epoch)	–	–	71.3%#24	Epoch AI Benchmarking Hub
SciCode	–	–	44.7%#87	SciCode
WeirdML	–	–	47.7%#72	WeirdML
GSO-Bench	–	–	14.7%#14	GSO-Bench
LMArena Coding	1520#29	–	1513#42	LMArena
LMArena WebDev	1393#73	–	1386#77	LMArena
SciCode (AA)not in index	–	45.7%#89	–	Artificial Analysis
LiveCodeBench	–	73.0%#80	–	Vals AI
IOI	–	18.3%#25	–	Vals AI
SWE-bench (Vals)not in index	–	70.0%#58	–	Vals AI
SWE-Bench Pro	–	–	43.6%#7	Scale AI SEAL
SWE-bench Verified (bash only)	71.4%#10	–	70.6%#12	SWE-bench
SWE-bench Verified (any scaffold)not in index	–	–	74.8%#9	SWE-bench

Agents & tools

Benchmark	high	thinking	default	Source
Terminal-Bench	–	–	46.5%#30	Terminal-Bench
GDPval	–	–	42.5%#4	OpenAI
Cybench	–	–	60.0%#3	Cybench
Remote Labor Index	–	–	2.1%#8	Scale AI / CAIS
Terminal-Bench Hard	–	35.6%#72	28.8%#115	Artificial Analysis
τ²-Bench Telecom (AA)not in index	–	78.1%#127	70.5%#150	Artificial Analysis
MCP Atlas	–	59.5%#24	–	Scale AI SEAL
BFCL Overall	–	–	73.2%#2	Berkeley Function Calling Leaderboard
τ²-bench	–	–	76.4%#7	τ²-bench

Maths

Benchmark	high	thinking	default	Source
FrontierMath Tiers 1–3	–	–	23.9%#68	Epoch AI Benchmarking Hub
FrontierMath Tier 4	–	–	2.4%#53	Epoch AI Benchmarking Hub
OTIS Mock AIME	–	–	77.8%#107	Epoch AI Benchmarking Hub
MATH Level 5	–	–	97.7%#6	Epoch AI Benchmarking Hub
ProofBench	–	–	19.0%#37	Vals AI
AIME (Vals)	–	88.2%#30	–	Vals AI
MGSM	–	94.3%#4	–	Vals AI
MathArena Apex	–	1.6%#33	–	MathArena

Knowledge

Benchmark	high	thinking	default	Source
SimpleQA Verified	–	–	30.7%#56	Epoch AI Benchmarking Hub
AA-Omniscience	–	-0.1#96	-9.4#135	Artificial Analysis
MMLU-Pro	–	87.4%#28	–	Vals AI
LegalBench	–	84.1%#35	–	Vals AI
CorpFin	–	62.0%#49	60.8%#63	Vals AI
TaxEval	–	73.3%#50	–	Vals AI
MedQA	–	94.7%#15	–	Vals AI
PRBench Finance	–	–	43.8%#19	Scale AI SEAL
PRBench Legal	–	–	40.8%#22	Scale AI SEAL
MultiNRC	–	35.8%#20	28.1%#25	Scale AI SEAL

Instruction following

Benchmark	high	thinking	default	Source
IFBench	–	57.3%#136	42.6%#231	Artificial Analysis
MultiChallenge	–	55.3%#19	–	Scale AI SEAL
TutorBench	–	49.0%#21	45.7%#24	Scale AI SEAL

Human preference

Benchmark	high	thinking	default	Source	Trend
LMArena Text	1456#57	–	1455#60	LMArena

Multimodal

Benchmark	high	thinking	default	Source
MMMU-Pro	–	68.7%#130	65.2%#144	Artificial Analysis
VISTA	–	48.8%#14	45.0%#30	Scale AI SEAL
MMMU (validation)	–	–	77.8%#11	MMMU
MMMU-Pro (official)not in index	–	–	68.9%#10	MMMU

Long context

Benchmark	high	thinking	default	Source	Trend
AA-LCR	–	72.3%#137	54.0%#239	Artificial Analysis

Composite

Benchmark	high	thinking	default	Source	Trend
Epoch Capabilities Indexnot in index	–	–	146.8#57	Epoch AI Benchmarking Hub
AA Intelligence Index	–	21.2#163	19.3#182	Artificial Analysis

Where to run it

Every provider serving this model through OpenRouter, with throughput and first-token latency measured on live traffic over the last 30 minutes and each provider’s own price. Purple marks the best in each column.

Provider	Speed	First token	Input $/M	Output $/M	Context	Quantisation
Anthropic	36 tok/s	1.59 s	$3.00	$15.00	1M	–
Google Vertex (Global)	36 tok/s	1.59 s	$3.00	$15.00	1M	–
Amazon Bedrock	35 tok/s	1.67 s	$3.00	$15.00	1M	–
Azure	33 tok/s	1.89 s	$3.00	$15.00	200k	–
Claude Platform on AWS	32 tok/s	1.50 s	$3.00	$15.00	1M	–

Price history

Listed price per 1M tokens over time, as recorded by OpenRouter for the provider with the longest history.

input outputnow $3.30 in · $16.50 out

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 at $0.300 per 1M. Reasoning tokens are not modelled.

Workload	Tokens in / out	Cost	With caching	Time
Chat reply	400 / 300	$0.0057	$0.0049	8.9 s
Summarise a 30-page report	12,000 / 600	$0.045	$0.021	16.4 s
Code edit	6,000 / 1,500	$0.041	$0.028	38.9 s
Agentic coding session	60,000 / 4,000	$0.240	$0.118	1.7 min
Structured extraction	2,000 / 200	$0.0090	$0.0050	6.4 s