Finally an Update from Anthropic (Claude 3.7)

Feb 25, 2025

Try the Requesty Router and get $6 free credits 🔀

Join the discord

We’re excited to unveil Claude 3.7 Sonnet, Anthropic’s most advanced model to date—a single, hybrid approach that seamlessly switches between near-instant responses and extended thinking for deep, step-by-step reasoning. This release brings major improvements in coding, web development, and general problem-solving. Alongside Claude 3.7 Sonnet, Anthropic is also introducing Claude Code, a command-line agentic coding tool in limited research preview.

Even better: Claude 3.7 Sonnet is already integrated with the Requesty Router—so you can start using it instantly in tools like Cline, Roo Code, OpenWebUI, and more. We’ve partnered with Anthropic to offer a discounted rate for a limited time, meaning you can test this new model at a lower cost.

Claude 3.7 Sonnet: Frontier Reasoning, Unified

Unlike other “reasoning models” that exist as separate variants, Claude 3.7 Sonnet merges fast response and deep reflection within a single LLM. You decide when it needs quick answers or extended reasoning—no separate model required.

Two Modes, One Model
- Standard Mode: Similar to a faster Claude, giving near-instant responses.
- Extended Thinking Mode: Claude 3.7 Sonnet “self-reflects” to tackle complex math, multi-step coding, physics, or intricate instruction-following.
Fine-Grained Control
- Through the API, you can limit how many “thinking tokens” are used, balancing speed and cost vs. more rigorous solutions.
Real-World Focus
- Anthropic has pivoted from purely competition benchmarks to tasks reflecting real developer and business needs—like front-end coding, agentic workflows, and robust debugging.

Performance Highlights

State-of-the-Art Coding
Independent tests show Claude 3.7 Sonnet leading in full-stack tasks, large codebase refactoring, and advanced tool use.
Agentic Workflows
On TAU-bench, which measures multi-step user and tool interactions, Claude 3.7 outperformed older Claude releases.
SWE-bench Verified
Reached 63.7% pass@1 on real software issues with minimal “scaffolding.” With a more advanced approach, it rose to 70.3%.

Claude Code (Limited Research Preview)

Claude Code is Anthropic’s new command-line assistant designed for end-to-end coding tasks. It can:

Search and read code in your file system or connected repo.
Edit files and run tests automatically.
Commit and push changes to GitHub.
Use CLI tools for debugging or building.

Think of it as a developer co-pilot that you can delegate big engineering tasks to. Anthropic found it shaved 45+ minutes off tasks that typically require manual building, refactoring, or debugging.

Interested in Claude Code? It’s available as a limited research preview. Sign up via Anthropic’s developer portal or the Requesty Router platform to request early access.

Using Claude 3.7 Sonnet in Your Workflow

Good news: the Requesty Router already hosts anthropic/claude-3-7-sonnet-latest, so you can integrate it right now with your favorite open-source and commercial tools.

1. Cline

Open Cline’s settings or UI.
Select Requesty as your provider.
Paste your Requesty API key.
Choose anthropic/claude-3-7-sonnet-latest as your model ID (or set an alias, e.g., alias/claude37).
Toggle extended thinking in Cline’s advanced settings to let Claude deeply reason or quickly respond.

2. Roo Code

In Roo Code, pick Requesty from the provider list.
Enter your API key.
Paste anthropic/claude-3-7-sonnet-latest as your model.
Enjoy improved code suggestions, multi-file refactoring, and advanced debugging right in your editor.

3. OpenWebUI

Open the Providers/Settings in OpenWebUI.
Set https://router.requesty.ai/v1 as the base endpoint.
Paste your API key.
Select “anthropic/claude-3-7-sonnet-latest” from the discovered models.
Interact with Claude via the OpenWebUI chat interface. Enable extended thinking with a single toggle (if supported).

4. Other Tools

We’re working with the creators of Leo AI (in Brave), Goose (for agentic tasks), Crew AI (collaborative brainstorming), Cursor (code editor), and more to ensure Claude 3.7 is ready to go. Just select or configure the anthropic/claude-3-7-sonnet-latest model in your tool’s settings.

Pricing & Availability

Same Prices as previous Claude versions:
- $3 per million input tokens
- $15 per million output tokens
Extended Thinking tokens also fall under the output token budget.
Limited-time Discount via Requesty:
- Enjoy a reduced price for the first 30 days. Check your Requesty dashboard for details.

Note: Extended thinking is not available on the free tier. Upgrade to Pro, Team, or Enterprise to unlock it.

Responsible Deployment

Anthropic has expanded safety measures in Claude 3.7 Sonnet:

Reduced Refusals by 45% for benign requests.
System Card details new security improvements, especially around prompt injection in agentic workflows.
Responsible Scaling: The model actively filters harmful queries and can interpret nuanced contexts accurately, leading to better user experiences.

Looking Ahead

Claude 3.7 Sonnet blends quick answers and deep reasoning in a single LLM, bridging the best of both worlds for developers. Combined with the new Claude Code, it makes an impressive leap toward fully agentic AI coding assistance. Anthropic is actively iterating on these features to handle larger tasks, more complex transformations, and even better reliability in the near future.

Join the Evolution

Try Claude 3.7 Sonnet in any Requesty-supported tool.
Enable extended thinking for your trickiest problems.
Test the preview of Claude Code if you have early access, or request an invite.
Send Feedback on performance, cost trade-offs, or UI/UX—directly shaping how Anthropic and Requesty refine these tools.

Have questions or looking for integration tips? Reach out on our Discord channel or check the Requesty docs for step-by-step setup guides. We can’t wait to see what you build with Claude 3.7 Sonnet—the new gold standard in unified AI reasoning.