Claude Sonnet 4.5

2025-09-29 · Publisher : Anthropic · Domain : Artificial Intelligence

Anthropic launches Claude Sonnet 4.5, presented as its best coding and agent model, able to stay focused for more than 30 hours. The Claude Agent SDK is released the same day.

What changes

  • 77.2% on SWE-bench Verified and 61.4% on OSWorld (computer use).
  • Announced autonomy of more than 30 hours on complex multi-step tasks.
  • Claude Agent SDK, on the same infrastructure as Claude Code, checkpoints in Claude Code and a native VS Code extension.
  • Context editing and a memory tool for long-running agents.
  • Model described as the most aligned released to date, with better resistance to prompt injection.

### Specifications - API identifier: claude-sonnet-4-5-20250929 (alias claude-sonnet-4-5). - Context 200K tokens, maximum output 64K tokens. Reliable knowledge cutoff: January 2025; training data up to July 2025 (documentation). - Manual extended thinking. Text and images as input, text as output. - Availability: Claude apps, API, Claude Code, Amazon Bedrock, Google Cloud Vertex AI. ### Pricing | Input (USD / M tokens) | Output (USD / M tokens) | Cache write 5 min / 1 h | Cache read | Batch | |---|---|---|---|---| | 3 | 15 | 3.75 / 6 | 0.30 | 1.50 / 7.50 | Price unchanged compared with Sonnet 4. ### Benchmarks | Benchmark | Score | |---|---| | SWE-bench Verified | 77.2% (200K context) | | SWE-bench Verified, 1M context configuration | 78.2% | | SWE-bench Verified, high compute | 82.0% | | OSWorld | 61.4% | Scores published by the vendor; the full comparison table is an image, not recorded. ### Key takeaways Sonnet 4.5 raises the range's coding performance at the price of Sonnet 4, and the Agent SDK opens up agent building on the same foundation as Claude Code.

Links

Tags : LLM, Anthropic, Claude, Sonnet, Agents, Codage, Agent SDK