claude-sonnet-4-5-20250929

  • 2025-09-29
  • Chat, Reasoning, Vision
  • By Anthropic
  • Top Tier
  • Input: ￥27.00 / M tokens Output: ￥135.00 / M tokens
  • Features: Function Calling, Image Input, Reasoning, Streaming, Structured Outputs, Text Input, Text Output, Web Search
  • Context Window: 1M
  • Maximum Output: 64K
  • Input: ￥27.00 / M tokens Output: ￥135.00 / M tokens
  • Features: Function Calling, Image Input, Reasoning, Streaming, Structured Outputs, Text Input, Text Output, Web Search
  • Context Window: 1M
  • Maximum Output: 64K

Model Description

JuheNext offers a total of four models:

  • claude-sonnet-4-5-20250929
  • claude-sonnet-4-5-20250929-thinking

Economical versions：

  • claudecode/claude-sonnet-4-5-20250929
  • claudecode/claude-sonnet-4-5-20250929-thinking

A Detailed Introduction to the New Model: Claude Sonnet 4.5

On September 30, 2025, Anthropic released Claude Sonnet 4.5, which the company officially calls “the world’s best programming model” and “the most powerful model to date.” This release introduces substantial improvements in performance, safety, and usability across the Claude product suite.

1. Main Features and Improvements

Claude Sonnet 4.5 demonstrates significant advancements in several key areas:

  • Outstanding Programming and Agent Capabilities: The model has achieved state-of-the-art performance on the SWE-bench Verified evaluation, a benchmark that measures real-world software coding abilities. Furthermore, it is capable of maintaining focus for over 30 hours on complex, multi-step tasks.

  • Revolutionary Computer Usage Abilities: Its score on the OSWorld benchmark, which tests AI models on real-world computer tasks, has seen a significant jump from 42.2% four and a half months ago to a leading 61.4%. The Claude for Chrome extension has been updated to fully leverage these capabilities, enabling the model to directly navigate browsers, fill spreadsheets, and complete various tasks.

  • Powerful Reasoning and Mathematical Abilities: Compared to previous models, including Opus 4.1, Sonnet 4.5 shows superior domain knowledge and reasoning skills in professional fields such as finance, law, medicine, and STEM.

  • Comprehensive Product Upgrades:

    • Claude Code: A highly anticipated “Checkpoints” feature has been added for developers, allowing them to save and roll back progress with a single click. The terminal interface has also been refreshed, and a native VS Code extension has been released.
    • Claude API: New context editing and memory tools have been introduced, enabling agents to run for longer durations and handle more complex tasks.
    • Claude Application: The application now integrates code execution and file creation (spreadsheets, slides, documents) directly within conversations.

  • Pricing: The pricing for Claude Sonnet 4.5 remains the same as Sonnet 4, at $3 per million input tokens and $15 per million output tokens.

2. Safety and Alignment

Anthropic describes Claude Sonnet 4.5 as the “most aligned frontier model to date.” It significantly reduces inappropriate behaviors such as flattery, deception, power-seeking, and encouraging delusions. The model has also made significant progress in resisting prompt injection attacks. It is released under AI Safety Level 3 (ASL-3) protection, which includes classifiers for chemical, biological, radiological, and nuclear weapons (CBRN) risks.

3. Developer Tools

Alongside the model, Anthropic has released the Claude Agent SDK. This provides developers with the core infrastructure that Anthropic uses to build its own products like Claude Code, allowing them to create their own AI agents for a wide range of tasks.

4. Research Preview

A limited-time research preview feature named “Imagine with Claude” has been launched. This feature allows Claude to generate software in real time, showcasing the potential of combining its advanced model capabilities with its infrastructure.

5. Customer Testimonials

The announcement is supported by high praise from experts and executives across various fields, including finance, law, healthcare, technology (GitHub, Canva, Figma), and security. These testimonials confirm Sonnet 4.5’s significant advances in handling complex tasks, code development, security analysis, and creative design.

Click to Buy

🔔How to Use

graph LR A("Purchase Now") --> B["Start Chat on Homepage"] A --> D["Read API Documentation"] B --> C["Register / Login"] C --> E["Enter Key"] D --> F["Enter Endpoint & Key"] E --> G("Start Using") F --> G style A fill:#f9f9f9,stroke:#333,stroke-width:1px style B fill:#f9f9f9,stroke:#333,stroke-width:1px style C fill:#f9f9f9,stroke:#333,stroke-width:1px style D fill:#f9f9f9,stroke:#333,stroke-width:1px style E fill:#f9f9f9,stroke:#333,stroke-width:1px style F fill:#f9f9f9,stroke:#333,stroke-width:1px style G fill:#f9f9f9,stroke:#333,stroke-width:1px

Purchase Now

Start Chat on Homepage

Register / Login

Enter Key

Read API Documentation

Enter Endpoint & Key

Start Using

Description Ends

Recommend Models

DeepClaude-3-7-sonnet

  • Chat, Reasoning
  • JuheAI
DeepSeek-R1 + claude-3-7-sonnet-20250219，The Deep series is composed of the DeepSeek-R1 (671b) model combined with the chain-of-thought reasoning of other models, fully utilizing the powerful capabilities of the DeepSeek chain-of-thought. It employs a strategy of leveraging other more powerful models for supplementation, thereby enhancing the overall model's capabilities.
  • 2025-02-19

DeepSeek-R1

  • Chat, Reasoning
  • DeepSeek
Performance on par with OpenAI-o1, Fully open-source model & technical report, Code and models are released under the MIT License: Distill & commercialize freely.
  • 2025-01-20

QwQ-32B

  • Chat, Reasoning
  • Qwen
QwQ-32B is a 32.5B-parameter reasoning model in the Qwen series, featuring advanced architecture and 131K-token context length, designed to outperform state-of-the-art models like DeepSeek-R1 in complex tasks.
  • 2025-03-06