- Claude Code on the web offers a browser-based platform for delegating coding tasks, enabling parallel execution across repositories with GitHub integration.
- The new Claude HighCoup 4.5 model provides fast, cost-effective code implementation, significantly outperforming its predecessor in speed and cost efficiency.
- Users can leverage a multi-agent workflow, using Claude Sonic 4.5 for planning and HighCoup 4.5 for execution, enhanced by features like multi-question prompting for vague requests.
Claude Code updates: When to use Haiku 4.5, Claude Code on web, and more.
- Claude Code on the web facilitates browser-based development, allowing users to connect GitHub repositories and delegate coding tasks to be executed in isolated sandboxes.
- It supports running multiple coding tasks simultaneously across different repositories, improving efficiency for bug backlogs or new feature development.
- The "open in CLI" feature enables users to seamlessly transfer AI-generated changes from the web environment to their local development setup while preserving full context.
- Claude HighCoup 4.5 is a new, highly efficient coding model that offers performance similar to Sonic 4 at one-third the cost and over twice the speed.
- Developers are encouraged to use a multi-agent approach, with Sonic 4.5 handling high-level problem decomposition and planning, and HighCoup 4.5 performing the actual code implementation.
- HighCoup 4.5 is priced competitively at $1 per million input tokens and $5 per million output tokens, making it suitable for large-scale operations.
- Claude Code now features multi-question support, allowing the AI to ask clarifying follow-up questions when presented with broad or vague prompts.
IDE — Integrated Development Environment; software that provides comprehensive facilities to computer programmers for software development.
Repo — Short for repository; a storage location for code, files, and revision history, commonly managed by version control systems like Git.
Input validation — The process of ensuring that user input meets specific criteria before being processed or stored.
Isolated sandbox — A secure, controlled environment that isolates applications or processes from the rest of the system, preventing malicious or buggy code from affecting other resources.
Agentic coding — A paradigm where an AI agent plans, executes, and iterates on coding tasks, often involving self-correction and multi-step reasoning.
Problem decomposition — The process of breaking down a complex problem into smaller, more manageable sub-problems.
Technical debt — The implied cost of additional rework caused by choosing an easy solution now instead of using a better approach that would take longer.
Tokens — The basic units of text that an AI model processes, often words or sub-word units, used for pricing and model capacity.
Caching — The process of storing copies of data in a temporary storage location (cache) so that future requests for that data can be served faster.
API — Application Programming Interface; a set of definitions and protocols for building and integrating application software, allowing different systems to communicate.
Lot code is now available on the web. So instead of loading up your terminal or having to open up your IDE anytime you want to make a change, you can now delegate coding tasks directly from your browser or the Claude mobile app. You simply connect your GitHub account, you select your repo, describe what you need done, and Claude handles the rest, all running on our Claude infrastructure. And the best part is you can now run multiple tasks in parallel across different repositories. So whether you have a bug backlog, routine fixes, or a brilliant new idea or feature you want to add, you can just spin up a session or multiple sessions and let Claude handle them simultaneously. Let me show you how this works. So I'll open up Claude Code on the web at claude.ai slash code. I'll select a repo. I have a repo already selected here, and then I'll give it a prompt. Let's say add input validation to the user registration endpoint. Also make sure we're using a strong password on both the one end and back end API. We'll hit enter and we are off the deraces. Claude is working in its own isolated sandbox with full-get access, and we can watch the progress in real time, but while this runs, let me tell you about some of the other stuff that we recently launched. The big one is Claude HighCoup 4.5. And a few weeks ago we launched CloudSonic 4.5, which became the world's best coding model. And just last week we followed it up with Claude HighCoup 4.5. And here's the really remarkable thing. Five months ago, Sonic 4 was state of the art. And now HighCoup 4.5 gives you a similar coding performance at one third of cost and more than twice the speed. It even surpasses Sonic 4 at certain tasks like computer use. Now the benchmarks are impressive, 73.3% on sweetbench verified, 41% on terminal bench, and it achieves 90% of Sonic 4.5's performance in agentic coding evaluations. But the big question you're probably having is when should developers use HighCoup 4.5? And the big pattern that we're seeing from our community is that Sonic 4.5 does the high level problem decomposition and planning. And then HighCoup 4.5 is used to actually implement and write the code. Now this multi-agent approach is becoming really popular for things like large scale refactors or tackling technical debt. HighCoup 4.5 is available wherever you get your Claude. And it's available at just $1 per million input tokens and $5 per million output tokens. Alright, let's check back on our task and see how it's doing. So if we open it up here, we can see that Claude gave us a good summary of all of the changes that it made in the service.go file, in the TypeScript on the front end. And my absolute favorite feature of Claude Code on the web is this open in CLI button. And the reason that I really like it is you simply click it, you open up your terminal, paste in the command that it gave you. And now I am working in my local development environment with all of the context and all of my progress from the web preserve. So I can see the messages, I can see all of the files that have been changed. I can go into VS Code and see all of the changes that were made to the service, sign up, assign up.touch script file and much more. And I can continue right from where I left off on my mobile phone or at the browser. Now I want to show you one final feature that we recently added to Claude Code. I gave it this prompt earlier, talking about adding, caching to my application, but I gave it a very broad prompt saying, hey, I would like to add caching to my application. And what Claude Code did is it's going to ask me a number of follow-up questions to make sure that it implements it correctly. So it's going to ask the caching level here, should there be database query caching, API response caching, and I can choose any of these, go to the next section, and figure out what my goals are, is it to improve performance, reduce database load. The target area is it for the workout data, for the user data, analytics, choose the technology, red is in memory, postgres, and then I can hit submit and have it go and implement based on the follow-up information that I gave it. And there you have it, Claude Code on the web for parallel cloud-based development at home and on the go, iKOO 4.5 for lightning fast coding at a fraction of the cost and smarter local workflows with multi-question support. Claude Code on the web is available now for pro and max users at claw.ai slash code. Give it a try and let us know what you think. Happy building!
TL;DR
- Claude Code on the web offers a browser-based platform for delegating coding tasks, enabling parallel execution across repositories with GitHub integration.
- The new Claude HighCoup 4.5 model provides fast, cost-effective code implementation, significantly outperforming its predecessor in speed and cost efficiency.
- Users can leverage a multi-agent workflow, using Claude Sonic 4.5 for planning and HighCoup 4.5 for execution, enhanced by features like multi-question prompting for vague requests.
Takeaways
- Claude Code on the web facilitates browser-based development, allowing users to connect GitHub repositories and delegate coding tasks to be executed in isolated sandboxes.
- It supports running multiple coding tasks simultaneously across different repositories, improving efficiency for bug backlogs or new feature development.
- The "open in CLI" feature enables users to seamlessly transfer AI-generated changes from the web environment to their local development setup while preserving full context.
- Claude HighCoup 4.5 is a new, highly efficient coding model that offers performance similar to Sonic 4 at one-third the cost and over twice the speed.
- Developers are encouraged to use a multi-agent approach, with Sonic 4.5 handling high-level problem decomposition and planning, and HighCoup 4.5 performing the actual code implementation.
- HighCoup 4.5 is priced competitively at $1 per million input tokens and $5 per million output tokens, making it suitable for large-scale operations.
- Claude Code now features multi-question support, allowing the AI to ask clarifying follow-up questions when presented with broad or vague prompts.
Vocabulary
IDE — Integrated Development Environment; software that provides comprehensive facilities to computer programmers for software development.
Repo — Short for repository; a storage location for code, files, and revision history, commonly managed by version control systems like Git.
Input validation — The process of ensuring that user input meets specific criteria before being processed or stored.
Isolated sandbox — A secure, controlled environment that isolates applications or processes from the rest of the system, preventing malicious or buggy code from affecting other resources.
Agentic coding — A paradigm where an AI agent plans, executes, and iterates on coding tasks, often involving self-correction and multi-step reasoning.
Problem decomposition — The process of breaking down a complex problem into smaller, more manageable sub-problems.
Technical debt — The implied cost of additional rework caused by choosing an easy solution now instead of using a better approach that would take longer.
Tokens — The basic units of text that an AI model processes, often words or sub-word units, used for pricing and model capacity.
Caching — The process of storing copies of data in a temporary storage location (cache) so that future requests for that data can be served faster.
API — Application Programming Interface; a set of definitions and protocols for building and integrating application software, allowing different systems to communicate.
Transcript
Lot code is now available on the web. So instead of loading up your terminal or having to open up your IDE anytime you want to make a change, you can now delegate coding tasks directly from your browser or the Claude mobile app. You simply connect your GitHub account, you select your repo, describe what you need done, and Claude handles the rest, all running on our Claude infrastructure. And the best part is you can now run multiple tasks in parallel across different repositories. So whether you have a bug backlog, routine fixes, or a brilliant new idea or feature you want to add, you can just spin up a session or multiple sessions and let Claude handle them simultaneously. Let me show you how this works. So I'll open up Claude Code on the web at claude.ai slash code. I'll select a repo. I have a repo already selected here, and then I'll give it a prompt. Let's say add input validation to the user registration endpoint. Also make sure we're using a strong password on both the one end and back end API. We'll hit enter and we are off the deraces. Claude is working in its own isolated sandbox with full-get access, and we can watch the progress in real time, but while this runs, let me tell you about some of the other stuff that we recently launched. The big one is Claude HighCoup 4.5. And a few weeks ago we launched CloudSonic 4.5, which became the world's best coding model. And just last week we followed it up with Claude HighCoup 4.5. And here's the really remarkable thing. Five months ago, Sonic 4 was state of the art. And now HighCoup 4.5 gives you a similar coding performance at one third of cost and more than twice the speed. It even surpasses Sonic 4 at certain tasks like computer use. Now the benchmarks are impressive, 73.3% on sweetbench verified, 41% on terminal bench, and it achieves 90% of Sonic 4.5's performance in agentic coding evaluations. But the big question you're probably having is when should developers use HighCoup 4.5? And the big pattern that we're seeing from our community is that Sonic 4.5 does the high level problem decomposition and planning. And then HighCoup 4.5 is used to actually implement and write the code. Now this multi-agent approach is becoming really popular for things like large scale refactors or tackling technical debt. HighCoup 4.5 is available wherever you get your Claude. And it's available at just $1 per million input tokens and $5 per million output tokens. Alright, let's check back on our task and see how it's doing. So if we open it up here, we can see that Claude gave us a good summary of all of the changes that it made in the service.go file, in the TypeScript on the front end. And my absolute favorite feature of Claude Code on the web is this open in CLI button. And the reason that I really like it is you simply click it, you open up your terminal, paste in the command that it gave you. And now I am working in my local development environment with all of the context and all of my progress from the web preserve. So I can see the messages, I can see all of the files that have been changed. I can go into VS Code and see all of the changes that were made to the service, sign up, assign up.touch script file and much more. And I can continue right from where I left off on my mobile phone or at the browser. Now I want to show you one final feature that we recently added to Claude Code. I gave it this prompt earlier, talking about adding, caching to my application, but I gave it a very broad prompt saying, hey, I would like to add caching to my application. And what Claude Code did is it's going to ask me a number of follow-up questions to make sure that it implements it correctly. So it's going to ask the caching level here, should there be database query caching, API response caching, and I can choose any of these, go to the next section, and figure out what my goals are, is it to improve performance, reduce database load. The target area is it for the workout data, for the user data, analytics, choose the technology, red is in memory, postgres, and then I can hit submit and have it go and implement based on the follow-up information that I gave it. And there you have it, Claude Code on the web for parallel cloud-based development at home and on the go, iKOO 4.5 for lightning fast coding at a fraction of the cost and smarter local workflows with multi-question support. Claude Code on the web is available now for pro and max users at claw.ai slash code. Give it a try and let us know what you think. Happy building!