Mix Hub AI
MixHubAI.com
Loading...

Anthropic Claude Sonnet 4.5

Explore Anthropic's latest release Claude Sonnet 4.5 — a breakthrough coding model with stronger benchmarks, longer task autonomy, and new developer tools.

2025年9月29日
4 min read
Claude SonnetAnthropicAI CodingLLMDeveloper Tools

Anthropic has just released Claude Sonnet 4.5, and they’re calling it the “best coding model in the world.” That’s a bold claim — but after looking at the benchmarks and new features, it doesn’t feel like marketing fluff. Instead, it seems like the model is genuinely pushing boundaries in how AI can help developers ship code, debug faster, and even build software on the fly.

👉 Want to try it yourself? Head over to the free Claude Sonnet 4.5 workspace.

What’s New in Claude Sonnet 4.5

1. Smarter, Longer Coding Sessions
Sonnet 4.5 can now run autonomously for 30 hours (Opus 4 tapped out at 7). That means it can handle longer projects without losing focus — useful for anyone working on complex multi-step builds.

2. Benchmarks Breaking Records

  • On SWE-Bench Verified, Sonnet 4.5 hit 77.2% (and 82% with test-time compute).
  • On OSWorld tasks, it leads the field at 61.4%, up dramatically from 43.9% in Sonnet 4.
  • In coding benchmarks, it continues to beat rivals like GPT‑5 and Gemini 2.5 Pro.

3. Practical Improvements

  • More reliable code refactoring.
  • Access to virtual machines + memory.
  • Multi-agent support for bigger workflows.

Developer Tools: More Than Just the Model

Anthropic isn’t just dropping a new model — they’re expanding the ecosystem.

  • Claude Code Updates: now with a VS Code extension, inline diffs, searchable history, and rollback checkpoints.
  • Claude Agent SDK: build your own agents with the same plumbing Anthropic uses. Includes orchestration, memory, permissions, and tool integration.
  • Imagine with Claude: experimental feature that lets you see apps being coded in real time. It’s limited to Claude Max subscribers for five days, but it hints at what “instant software” might look like in practice.

Why This Release Matters

For developers drowning in repetitive fixes and debugging, Claude Sonnet 4.5 feels like a lifeline:

  • You spend less time nudging the model back on track.
  • Long tasks don’t collapse partway through.
  • The outputs don’t just “look like” code — they actually compile and solve the job.

Cursor’s CEO Michael Truell said it best: “State-of-the-art coding performance with significant improvements on longer horizon tasks.”

Pricing

Pricing for Sonnet 4.5 will remain at $3/$15 per million tokens of input/output, the same as Anthropic previously charged for Sonnet 4.

My Take

Claude Sonnet 4.5 isn’t just another incremental upgrade. It’s a noticeable leap in autonomy, reliability, and coding accuracy. And while benchmarks are impressive, what excites me more is the ecosystem: the SDK, the VS Code integration, and even playful experiments like “Imagine with Claude.” Those tools are where everyday dev workflows start to evolve.

If coding with AI felt like “pair programming with training wheels” before, Claude’s new release feels closer to working with an actual senior engineer who doesn’t get tired after 12 hours. And that is a big deal.

👉 Try it now: Claude Sonnet 4.5 free.