Coding, agents, safety: all you need to know about Claude Sonnet 4.5

by
New Delhi: Anthropic has rolled out Claude Sonnet 4.5, and the pitch is big. The company says this model can handle serious coding work, run for long stretches, and use a computer like a careful assistant. We spent the morning going through the docs and updates. The thing that stood out to us is how much of this release is about real work, not just flashy demos.

Anthropic is calling it a leap in coding, agent building, and safe behavior. The blog opens with a bold line. “Claude Sonnet 4.5 is the best coding model in the world. It’s the strongest model for building complex agents. It’s the best model at using computers. And it shows substantial gains in reasoning and math.” That sets the tone for everything else.

What is Claude Sonnet 4.5

Anthropic says Sonnet 4.5 leads on SWE-bench Verified for real software tasks, and hits 61.4 percent on OSWorld for computer use. The company also claims longer focus. In its words, the model has “maintaining focus for more than 30 hours on complex, multi-step tasks.” In plain terms, it keeps working without losing the thread.

Alongside the model, Anthropic refreshed Claude Code with checkpoints and a cleaner terminal. The apps now let you execute code and make files like spreadsheets and slides right inside the chat. I like that small touch because it cuts the constant app-switching we all do.

Claude Sonnet 4.5 Benchmarks | SOurce; Anthropic

Key upgrades for developers

  • Checkpoints in Claude Code to save and roll back work
  • Native VS Code extension and a new terminal
  • Context editing and a memory tool in the API for long jobs
  • Better tool use with parallel calls and smarter context handling

Anthropic adds, “This is the most aligned frontier model we’ve ever released,” pointing to fewer issues like sycophancy, deception, and power seeking. The system card covers the safety tests and the ASL-3 guardrails.

Claude Agent SDK

Anthropic is shipping the same building blocks it used for Claude Code. “We’re calling this the Claude Agent SDK.” The idea is simple. You get the plumbing to build agents that run longer, manage memory, ask permissions, and coordinate sub-agents. If you are building research bots, finance helpers, or cybersec tools, this is the bit you will likely test first.

Pricing and availability in India terms

Pricing stays the same as Sonnet 4. At $3 per million input tokens and $15 per million output tokens, that is roughly ₹261 for input and about ₹1,305 for output at 1 USD equal to ₹87. Sonnet 4.5 is live on the Claude API and also on Amazon Bedrock and Google Cloud Vertex AI. It is available in the Claude apps and Claude Code too.

One more thing

There is a short research preview called “Imagine with Claude.” It shows the model generating software in real time. No preset code. No fixed flows. It is available to Max users for a few days. Think of it as a peek into where agentic tools might go next.