On June 30, 2026, Anthropic launched Claude Sonnet 5 – and it's not an incremental update. The model is designed to be able to plan, use tools such as browsers and terminals, and drive autonomously with minimal human control. Anthropic calls it "the most agentic Sonnet model yet."
Near Opus class - at Sonnet price
What stands out is the performance gap compared to its predecessor. Claude Sonnet 5 is close to Claude Opus 4.8 in benchmarks, but costs like a Sonnet. The improvements are greatest in:
- Coding – write, review and debug code more reliably
- Reasoning – handles multi-step problems and logic chains better
- Tool usage – able to coordinate external tools and APIs as part of an autonomous flow
- Knowledge work – better at summarizing, analyzing and drawing conclusions from complex information
The model also supports different effort levels, allowing you to scale cost versus capacity depending on what the task requires.
Safer and more reliable
Sonnet 5 shows a lower incidence of unwanted behaviors compared to Sonnet 4.6. It is better at refusing malicious requests, more resistant to prompt injection attacks, and hallucinates less. Security features are enabled by default.
Price and availability
Claude Sonnet 5 is the default model for the Free and Pro plans on Claude.ai and is available for Max, Team and Enterprise users. Via API it is open to everyone.
| Period | Input | Output |
|---|---|---|
| Intro (until 31 Aug 2026) | $2 / 1M tokens | $10 / 1M tokens |
| Standard (from 1 Sep 2026) | $3 / 1M tokens | $15 / 1M tokens |
The model is also available via AWS Bedrock, Google Cloud Vertex AI, and Microsoft Foundry.
Source: Anthropic – Claude Sonnet 5
