Claude Opus 4.6 is Anthropic's most intelligent model released in February 2026, built for agents that operate across entire workflows rather than single prompts. It features a 1M-token context window, 128K max output tokens, and the ability to spawn and coordinate multiple sub-agents working in parallel — a capability called Agent Teams. With adaptive thinking that dynamically adjusts reasoning depth, the model excels at large codebases, complex refactors, sustained knowledge work, and end-to-end project execution, producing near-production-ready documents and analyses in a single pass.
Anthropic ProAnthropic Max (5x)Anthropic Max (20x)API|VisionReasoningWeb Search|Proprietary Model
Knowledge Cutoff
2025-09-01
Input → Output Format
Context Memory
1MIN128KOUT
AI Performance Evaluation
Arena Overall Score
1503
±5As of 2026-04-23
Overall Rank
🥈 No.2
20,192 Votes
Arena by Ability
Hard Prompts
1535±6🥇 No.1
Expert Knowledge
1544±16🥈 No.2
Instruction Following
1516±9🥇 No.1
Conversation Memory
1514±11🥉 No.3
Creative
1494±11🥈 No.2
Coding
1554±9🥉 No.3
Math
1517±17🥇 No.1
Arena by Occupation
Creative Writing
1494±9🥈 No.2
Social Sciences
1517±11🥈 No.2
Media
1485±10🥇 No.1
Business
1499±10🥈 No.2
Healthcare
1512±16No.7
Legal
1514±16🥇 No.1
Software
1541±8🥈 No.2
Mathematics
1520±20🥈 No.2
Source:Arena Intelligence
Overall
AA Intelligence Index
53%↑15%
LiveBench
77%↑17%
ForecastBench
59%↑0%
Reasoning & Math
GPQA Diamond
90%↑9%
HLE
37%↑20%
LB Reasoning
89%↑29%
LB Math
89%↑16%
LB Data
70%↑20%
Coding
AA Coding Index
48%↑14%
LB Coding
78%↑5%
LB Agentic
62%↑18%
TAU2
92%↑19%
TerminalBench
46%↑15%
SciCode
52%↑11%
Language & Instructions
IFBench
53%↓4%
AA-LCR
71%↑9%
Hallucination (HHEM)
12%↑2%
Factual (HHEM)
88%↓2%
LB Language
83%↑11%
LB IF
63%↑17%
Output Speed
Standard Mode
45tok/s↓37
First Output 1.75s
Reasoning Mode
58tok/s↓30
First Output 12.62s
