Claude Sonnet 4.5 is Anthropic's most advanced Sonnet model prior to the 4.6 generation, optimized for real-world agents and coding workflows. It delivers state-of-the-art coding performance with improvements in system design, code security, and specification adherence. The model is designed for extended autonomous operation with fact-based progress tracking, and introduces stronger agentic capabilities including speculative parallel execution, improved tool orchestration, and efficient context and memory management for multi-context, long-running workflows.
Anthropic FreeAnthropic ProAnthropic Max (5x)Anthropic Max (20x)API|VisionReasoningWeb SearchFile|Proprietary Model
Knowledge Cutoff
2025-01-31
Input → Output Format
Context Memory
1MIN64KOUT
AI Performance Evaluation
Arena Overall Score
1454
±3As of 2026-05-01
Overall Rank
No.36
65,936 Votes
Arena by Ability
Hard Prompts
1485±4No.25
Expert Knowledge
1496±10No.17
Instruction Following
1464±5No.16
Conversation Memory
1470±6No.30
Creative
1443±7No.24
Coding
1520±6No.13
Math
1455±9No.33
Arena by Occupation
Creative Writing
1447±6No.29
Social Sciences
1468±6No.39
Media
1440±6No.22
Business
1455±6No.32
Healthcare
1466±10No.51
Legal
1458±9No.44
Software
1499±5No.23
Mathematics
1469±11No.25
Source:Arena Intelligence
Overall
AA Intelligence Index
43%↑4%
LiveBench
51%↓9%
ForecastBench
60%↑1%
Reasoning & Math
AA Math Index
88%↑14%
GPQA Diamond
83%↑1%
HLE
17%↑0%
MMLU-Pro
88%↑6%
AIME 2025
88%↑14%
LB Reasoning
42%↓27%
LB Math
63%↓12%
LB Data
47%↓6%
Coding
AA Coding Index
39%↑2%
LiveCodeBench
71%↑6%
LB Coding
76%↑3%
LB Agentic
48%↑3%
TAU2
78%↓2%
TerminalBench
36%↑2%
SciCode
45%↑3%
Language & Instructions
IFBench
57%↓6%
AA-LCR
66%↑4%
Hallucination (HHEM)
10%↑0%
Factual (HHEM)
90%↑0%
LB Language
76%↑4%
LB IF
24%↓27%
Output Speed
Standard Mode
44tok/s↓33
First Output 0.80s
Reasoning Mode
46tok/s↓41
First Output 8.97s
