Anthropic has launched Claude Sonnet 5, its most agent-focused Sonnet model to date, with improved performance in autonomous work, coding, reasoning, and tool use.
Anthropic said the new model performs close to Claude Opus 4.8 while costing less.
The model can create plans, operate tools such as web browsers and terminals, and complete multi-step tasks with minimal human intervention. Anthropic said that only a few months ago, this level of autonomous performance required larger, more expensive models.
Claude Sonnet 5 narrows the performance gap between Anthropic’s Sonnet and Opus model families. It also improves on Claude Sonnet 4.6 in reasoning, coding, tool use, and professional knowledge work.
The model can follow plans across longer tasks, select appropriate tools, correct errors, and continue working through multiple steps. Anthropic has positioned it for software development, browser-based tasks, internal automation, and customer-facing AI agents.
At higher effort settings, Claude Sonnet 5 can match Claude Opus 4.8 on some agentic search and computer-use tasks. Developers can adjust the effort level to balance performance, speed, and operating costs.
Claude Sonnet 5 supports a context window of 1 million tokens and can generate up to 128,000 output tokens in a standard synchronous API request.
It also supports adaptive thinking, allowing the model to adjust how much reasoning it uses according to the difficulty of a task.
Developers can access the model through the Claude API using the claude-sonnet-5 model identifier. It is also available through Amazon Web Services, Google Cloud, and Microsoft Foundry.
Claude Sonnet 5 is available across all Claude subscription plans. It is now the default model for Free and Pro users, while Max, Team, and Enterprise subscribers can also select it. The model is available in Claude Code and on the Claude Platform.
Anthropic is charging an introductory API price of $2 per million input tokens and $10 per million output tokens until August 31, 2026.
Standard pricing will increase to $3 per million input tokens and $15 per million output tokens from September 1, 2026. Prompt caching can reduce costs by up to 90%, while batch processing can reduce costs by up to 50%.
Claude Sonnet 5 uses an updated tokenizer that can convert the same content into roughly 1 to 1.35 times as many tokens, depending on the type of material. Anthropic said it set the temporary introductory price to keep migration costs broadly similar to Claude Sonnet 4.6.
Anthropic’s tests found that Claude Sonnet 5 produced fewer undesirable behaviours than Sonnet 4.6.
The model was better at rejecting harmful requests and resisting prompt-injection attempts. It also recorded lower hallucination and sycophancy rates than its predecessor, although it showed more misaligned behaviour than Claude Opus 4.8 and Mythos Preview in Anthropic’s automated evaluation.
Anthropic did not specifically train Sonnet 5 for advanced cybersecurity work. The model performed considerably worse than the company’s Opus models on tests involving potentially dangerous cyber capabilities and did not produce a complete working exploit during the reported evaluation.
The company has still enabled real-time cybersecurity safeguards by default because Sonnet 5 is slightly more capable than Sonnet 4.6 in this area.
Get the latest tech news, telecom insights, and product launches wherever you prefer.
Add ProPakistani to Preferred Sources and see more of our stories in Google Search and Top Stories.
