Claude Sonnet 4.5 Release News and Improvements
- Ethan Carter
- 3 days ago
- 9 min read
Claude Sonnet 4.5 came out on September 9, 2025. It has an ASL-3 safety level. You can now use the Code Interpreter. This tool helps you with coding and automates tasks. The model can run for up to 30 hours by itself. This helps with bigger projects. You can use it on Amazon Bedrock. You can also try it in GitHub Copilot.
Claude AI has 3.9% of the total AI chatbot market. It has 29% of the enterprise AI market. This shows many businesses use it.
Feature | Impact on Productivity |
Makes coding faster and helps automate tasks. | |
Context Processing | Helps you decide and finish tasks more quickly. |
Key Takeaways
Claude Sonnet 4.5 has a new Code Interpreter. It helps you code faster and do tasks automatically. This makes it great for hard projects. The model now works by itself for over 30 hours. You can finish bigger jobs without stopping. Claude Sonnet 4.5 has an ASL-3 safety level. It can do harder tasks, but you need to watch it carefully.
You can use Claude Sonnet 4.5 on many platforms. These include Amazon Bedrock and GitHub Copilot. This makes it useful for many needs. The update makes coding much better. It gets higher scores on tests. This helps developers trust it more.
Release News
Release Date
You can write down the release date for Claude Sonnet 4.5. It came out on September 29, 2025. Anthropic shared the news on their website. They gave a full introduction there. This update has many changes you will see quickly.
Safety Level
It is important to know the safety level for this update. Claude Sonnet 4.5 uses the ASL-3 safety level now. This level lets the model do harder tasks. It also needs better safety steps. ASL-3 is higher than ASL-2, which was about safe actions. With ASL-3, you get more choices and power. You must be careful when you use the model.
Safety Level | Description | Key Characteristics |
ASL-2 | Bigger and harder AI models need stronger safety rules but are still mostly easy to control and predict. | Stronger safety rules, easy to control actions |
ASL-3 | Shows much more risk with actions that are harder to predict, so it needs smart safety steps. | Handles hard tasks, more chance for wrong use |
ASL-4 | Means even more risk with models that make choices by themselves. | Needs advanced ways to understand, could cause harm |
Tip: Always check your settings and watch how you use the model when working with ASL-3 or higher.
Availability
There are many ways to use Claude Sonnet 4.5. You can pick the platform that works best for you.
On Amazon Bedrock, you can use Claude Sonnet 4.5 if you build agents. The update is ready now. You do not need extra approval.
If you use GitHub Copilot, you can try new features in public preview. You need Copilot Pro, Pro+, Business, or Enterprise. An admin must turn on the Claude Sonnet 4.5 policy in your settings.
You can also use the model on Claude.ai. It works on web, iOS, and Android. You can use it almost anywhere.
Platform | Access Requirements |
GitHub Copilot | Works for Copilot Pro, Pro+, Business, and Enterprise. Admins must turn on the Claude Sonnet 4.5 policy in settings. |
Amazon Bedrock | Ready for developers who want to build agents. No special access steps listed. |
Anyone can use it on web, iOS, and Android. |
You will see that Claude Sonnet 4.5 on Amazon Bedrock is special. This platform lets you build and test agents with new updates. The update adds things like the Code Interpreter and longer running time. These help you finish tasks faster and work on harder projects. The model can run by itself for up to 30 hours. You can trust it with bigger jobs.
Feature | Description |
Speed | Claude Sonnet 4.5 is about 50% faster than older versions, so you finish tasks like code reviews much quicker. |
Performance | The model is great for long jobs, like making documents and fixing tough bugs. |
Code Interpreter | This new tool helps you work with many files and coding tasks easily. |
Extended Autonomous Operation | The model does agent tasks well and fast, like coding and making documents. |
Note: Using Claude Sonnet 4.5 on Amazon Bedrock gives you early access to new updates and features. You can start building smarter agents right away.
Claude Sonnet 4.5 Features & Main Improvements
Coding Upgrades
You will notice big changes in coding with Claude Sonnet 4.5. The new model brings a Code Interpreter that helps you write, test, and fix code faster. Many users say this update makes coding smoother and more reliable. Benchmarks show Claude Sonnet 4.5 leads in math, reasoning, and agent-building. It scored 61.4% on the OSWorld benchmark, which is almost 20% higher than before. You can now handle complex tasks like reading graphs and analyzing financial data with more confidence.
Claude Sonnet 4.5 is now the best coding model for many users.
It works well for long coding sessions and big projects.
Some users feel unsure about the new model, but most feedback is positive.
You can use these coding upgrades on Amazon Bedrock, which makes it easy to build and test your own agents.
Tip: Try the Code Interpreter for tasks that need careful math or logic. You will see better results, especially in finance and cybersecurity.
Claude Sonnet 4.5 also helps you find and fix security problems in your code. The new model can patch vulnerabilities and run deep financial checks. This means you can trust it for both business and personal projects.
Autonomous Operation
With this release, you get much longer autonomous operation. Claude Sonnet 4.5 can now work by itself for over 30 hours. This is a huge jump from the previous 7-hour limit in Claude Opus 4. You can let the new model handle big jobs, like writing documents or reviewing code, without stopping.
Feature | Claude Sonnet 4.5 | Claude Opus 4 | Improvement |
Maximum Autonomous Duration | Over 30 hours | 7 hours | Significant |
OSWorld Score | 61.4% | 42.2% | Major |
SWE-bench Verified Score | 77.2% | N/A | Significant |
You can use this extended operation on Amazon Bedrock. The platform supports long-running agents, so you do not need to restart your tasks often. This helps you finish bigger projects with less effort. The new model also works well for agent tasks, like coding, document creation, and data analysis.
Usage Limits
Claude Sonnet 4.5 gives you more hours to use each week. You can see your usage limits clearly, which helps you plan your work. The new features let you choose the best plan for your needs. Here is how the weekly hours compare for different subscription plans:
Subscription Plan | Weekly Hours of Sonnet 4 | Weekly Hours of Opus 4 |
Pro ($20/month) | 40 to 80 hours | N/A |
Max ($100/month) | 140 to 280 hours | 15 to 35 hours |
Max ($200/month) | 240 to 480 hours | 24 to 40 hours |
You can access these new features on Amazon Bedrock. The platform lets you track your usage and upgrade your plan if you need more hours. This update makes it easier to manage your time and resources.
Note: You can use Claude Sonnet 4.5 on Amazon Bedrock, GitHub Copilot, and other platforms. The new model brings updates that help you work faster and safer.
Comparison
Claude Sonnet 4.5 vs Previous Versions
When you look at Claude Sonnet 4.5 and older models, you see big changes. The new version lets you work longer without stopping. It gets better scores and has new tools. You can do more agent tasks and rebuild web apps easily. The table below shows how the main features got better:
Feature/Metric | Claude Sonnet 4.5 | Previous Versions (e.g., Sonnet 4) | Improvement |
Autonomy Duration | 30 hours | 7 hours | Significant |
Success Rate on OSWorld Benchmark | 61.4% | 43.9% | Major leap |
SWE-Bench Verified Score | 77.2% (82% with parallel compute) | N/A | High score |
Context Management | Improved | N/A | Enhanced |
Multi-Agent Support | Added | N/A | New feature |
Ability to Rebuild Web App | Yes | No | New capability |
These updates give you more control and better results. You can finish harder jobs and trust the model for longer work.
Claude Sonnet 4.5 vs Competitors
You might wonder how Claude Sonnet 4.5 compares to other top AI models. The new version has a bigger context window and newer knowledge. You can handle more information at once and get answers from recent data. The table below shows the main differences:
Feature | GPT-4.5 | Claude Sonnet 4.5 |
Input Context Window | 128K tokens | |
Maximum Output Tokens | 16.4K tokens | 128K tokens |
Knowledge Cut-off Date | October 2023 | April 2025 |
You also get strong coding and reasoning skills. Gemini Pro 2.5 has a huge context window and works well with Google tools. GPT-4.5 is good for chatting and has many plugins. Claude Sonnet 4.5 stands out with agent support and coding upgrades.
GPT-4.5: Good chat, lots of plugins, strong context window.
Claude Sonnet 4.5: Biggest context window, best for agent tasks, newest updates.
Gemini Pro 2.5: Works with Google, supports many types of input, very large context window.
Pick the model that works best for you. If you want longer tasks and better coding, Claude Sonnet 4.5 gives you more choices.
Tests show Claude Sonnet 4.5 gets high scores on SWE-bench and GPQA Diamond. You get strong performance for coding and reasoning. You can trust the model for business and developer work.
Use Cases
Business
Claude Sonnet 4.5 helps many businesses solve real problems. In hotels, it makes guest check-in faster and turns feedback into useful ideas. In the energy field, it quickly looks at reports and safety papers. Stores use it to match invoices and help customers better. Government teams use it to speed up permits and benefits. The table below shows how different jobs use the model and what they get from it:
Industry | Application Description | Reported Results |
Hospitality | Instant guest check-in, automated ID and booking confirmation | Fewer front-desk errors, faster feedback analysis |
Energy | Processes field reports, maintenance logs, safety documents | Faster compliance, lower audit risk, shorter billing cycles |
Retail | Automated supplier invoice reconciliation | Faster customer service, real-time insights |
Public Sector | Faster permit and benefit application processing | More accurate tax processing, lighter staff workload |
You can use Amazon Bedrock to make agents that do these jobs and watch progress as it happens. This helps you keep track of work, make fewer mistakes, and choose better options.
Developers
You get lots of tools that help you code faster and better. The model lets you make software right away and run code in virtual machines. You can handle long jobs with better memory and context. Teams can work together on hard projects with multi-agent support. You can also undo changes with checkpoints, so you make fewer mistakes.
Developers say coding is easier and more correct with this model.
You can fix code, add new things, and edit with less trouble.
The model remembers your work, even if you work for a long time.
Feature | Benefit for Developers |
Performance Metrics | Top coding performance, better than other models |
Access to Virtual Machines | Test and run code safely |
Multi-Agent Support | Teams can work together on big projects |
Checkpoints | Roll back changes easily |
Claude Agent SDK | Build custom agents for your needs |
Enhanced Tool Orchestration | Use many tools together for better workflows |
You can use amazon bedrock to get these tools and make your own coding agents.
Integration
You can add the model to your work using amazon bedrock or GitHub Copilot. On amazon bedrock, pick the model and set up your agent. You can watch how much you use it, set limits, and upgrade your plan if needed. For GitHub Copilot, turn on the model in admin settings if you have the right account.
Platform | Integration Steps |
Amazon Bedrock | Select the model, set up your agent, monitor usage, and manage plans |
GitHub Copilot | Enable in admin settings for eligible accounts, access advanced coding features |
Tip: Use amazon bedrock to try new features first and build agents that fit your work or coding needs.
You can also use the model to help with cybersecurity. It helps you do hard jobs and keeps your data safe. The model is good for money planning and guessing future trends, and it helps stop leaks of information.
You get more benefits with Claude Sonnet 4.5. The new terminal interface makes your work faster. Checkpoints help you fix mistakes quickly. Better coding tools keep your work safe. People say their code is more accurate now. Agents are smarter and make fewer mistakes. Businesses finish contract reviews faster. Cybersecurity is stronger for companies.To begin, change your model name. Keep using your current API calls. Turn on new features like memory tools for agents that run a long time.
FAQ
How do you enable Claude Sonnet 4.5 in GitHub Copilot?
You need an eligible Copilot account. Go to admin settings. Turn on the Claude Sonnet 4.5 policy. You can start using new coding features right away.
What is the ASL-3 safety level?
ASL-3 lets you use advanced AI tasks. You must follow safety steps. The model can handle harder jobs, but you should monitor its actions closely.
Can you use Claude Sonnet 4.5 for long projects?
Yes, you can run agent tasks for up to 30 hours. This helps you finish big jobs like code reviews or document creation without stopping.
Where can you track your usage limits?
You can check your usage on Amazon Bedrock. The dashboard shows your hours and lets you upgrade your plan if you need more time.
