Claude Sonnet 4.6

Overview

Claude Sonnet 4.6 is a major upgrade to Anthropic's Sonnet series of models, designed to deliver performance close to its flagship Opus model at a more economical cost. It achieves comprehensive improvements in coding, computer use, long-context reasoning, and knowledge work, and for the first time in the Sonnet series, introduces a 1,000,000 token context window (in beta), making it an ideal choice for enterprise-level automation and complex task processing.

Key Features

Near-Opus Performance: In multiple benchmarks, especially in the OfficeQA test for processing enterprise documents (such as charts, PDFs, tables), Sonnet 4.6's performance is comparable to Opus 4.6 [1].
1M Token Context Window: Brings a million-token context window to the Sonnet series for the first time, capable of processing entire codebases or massive research papers, greatly enhancing its long-range reasoning and analysis capabilities [2].
Excellent Coding and Computer Use Capabilities: Performs outstandingly on OSWorld (a benchmark measuring AI's ability to use real software) and has been confirmed by early users to reach or exceed human-level performance in complex code fixes, front-end page building, and multi-step web form filling.
Distinction from 4.5 Version: Sonnet 4.6 is a comprehensive skill upgrade over 4.5. It not only scores higher on various benchmarks but, more importantly, demonstrates stronger instruction-following, fewer hallucinations, and more reliable multi-step task execution in practical applications. Its performance improvement is so significant that many tasks that previously required an Opus model can now be accomplished with the more cost-effective Sonnet 4.6.

Best Use Cases

Enterprise-level Document Understanding and Analysis: Efficiently processes and reasons with complex PDF documents containing charts and tables.
Large-Scale Codebase Coding and Refactoring: With its million-token context window, it is ideal for codebase-level analysis and modification.
Cost-Effective Agentic Workflows: Provides near-top-tier performance and better cost-effectiveness in automated tasks requiring multi-step planning and execution.
Knowledge Work and Design: Excels in knowledge Q&A, content generation, and front-end design, with high-quality output and fewer iterations required.

Capabilities and Limitations

Capability	Detailed Description
Reasoning Ability	Strong. Its performance in long-context reasoning and complex problem-solving is close to the Opus level.
Creative Ability	Strong. Particularly adept at generating high-quality code and front-end pages with a good sense of design.
Multimodal Ability	Supported. Capable of processing and understanding multimodal inputs such as images and PDFs.
Response Speed	Medium. Achieves an excellent balance between performance and speed.
Context Window	1,000,000 Tokens (beta)
Max Output	128,000 Tokens [3]

Credits and Pricing

Model	Input (Credits/Token)	Output (Credits/Token)
Claude Sonnet 4.6	3.00	15.00