Anthropic announced the launch of Claude Sonnet 5. Claude Sonnet 5 is designed to be Anthropic’s most agentic Sonnet model yet. The company said the model can make plans, use tools such as browsers and terminals, and run autonomously at a level that previously required larger, more expensive models.
Anthropic said Sonnet 5 narrows the performance gap with Opus-class models while offering lower pricing. The model shows substantial improvement over Claude Sonnet 4.6 in areas including reasoning, tool use, coding, and knowledge work. Anthropic said Sonnet 5’s performance is close to Claude Opus 4.8 in certain areas while offering a wider range of cost-performance options. Claude Sonnet 5 is available across all plans.
It is the default model for Free and Pro plans and is also available to Max, Team, and Enterprise users. The model is also available in Claude Code and on the Claude Platform.
Developers can access the model through the Claude API using claude-sonnet-5. Claude Sonnet 5 is launching with introductory pricing of $2 per million input tokens and $10 per million output tokens through August 31, 2026.
After that period, pricing will move to $3 per million input tokens and $15 per million output tokens. Anthropic said users can adjust effort levels to balance cost and performance across different projects.
The company said early access partners found Sonnet 5 to be more agentic than previous Sonnet models.
Testers said the model can complete complex tasks that earlier Sonnet models would stop short of, check its own output without being asked, and perform agentic work at an attractive price point.
Anthropic also reported that its safety assessments found Sonnet 5 had a lower overall rate of undesirable behaviors than Sonnet 4.6. The company said Sonnet 5 is generally safer to use in agentic contexts and better at rejecting malicious requests and resisting hijack attempts during prompt injection attacks.
Anthropic also said the model shows lower rates of hallucination and sycophancy than Sonnet 4.6. However, Sonnet 5 showed somewhat higher rates of misaligned behavior than Opus 4.8 and Claude Mythos Preview in Anthropic’s automated behavioral audit. Anthropic said it did not deliberately train Sonnet 5 on cybersecurity tasks.
The model can perform some routine, non-harmful cyber tasks, but the company said it has a much lower ability to perform dangerous cybersecurity tasks than current Opus models. Sonnet 5 has been launched with cyber safeguards enabled by default.
Anthropic said those safeguards detect and block dangerous cyber usage in real time. The company also published a full Claude Sonnet 5 System Card with additional safety and capability evaluations.
Anthropic said the model uses an updated tokenizer, which can affect how text maps to tokens. The company said introductory pricing is designed to make the transition to Sonnet 5 roughly cost-neutral.
