Claude Sonnet 4.5

by Anthropic

Frontier foundation model optimized for agents, coding, and long-running tasks (Sonnet 4.5)

Part of the Anthropic Claude 4 series

See Anthropic release notes (Sept 29, 2025)

Summary

Sonnet 4.5 is Anthropic’s 2025 model release focused on safety, agentic workflows, and developer productivity. It’s positioned as a top coding and agent model with very large context capacity and improvements in alignment and safety.

Features

  • Released: September 29, 2025
  • Very large context window (reported ~200k tokens)
  • Optimized for coding, agentic automation, and “computer use” tasks
  • Strong domain knowledge improvements (coding, finance, cybersecurity)
  • Built-in context-awareness and token-usage tracking (usage visibility during tool calls)
  • Deployment across Claude.ai, Anthropic Developer Platform, Amazon Bedrock, Google Cloud (Vertex AI)
  • Pricing (announced): ~15 per million output tokens (check current docs)

Superpowers

Sonnet 4.5 is best for teams building complex agents or doing deep-codebase work: long-running tasks, patching, multi-file refactors, and building autonomous agents. It reduces friction when working across large contexts and retains alignment improvements to lower safety risk in sensitive scenarios.

Pricing

  • Published pricing example: 15 /M output tokens (confirm with Anthropic portal as discounts/promotions/enterprise pricing may apply).

Safety & alignment

  • Improved safety profile vs prior models (reductions in harmful behaviors in internal evals).
  • Better handling of adversarial scenarios (blackmail, framing, sabotage) per Anthropic system card and evals.
  • Designed for enterprise and internal deployment with strong monitoring and tooling.

Sources / notes:

  • Anthropic Sonnet 4.5 release notes, system card, and third-party summaries.