Anthropic launches Claude Sonnet 5 as a cheaper, more capable agentic AI model

Anthropic has released Claude Sonnet 5, the latest version of its mid-size AI model, positioning it as a more affordable and agentic alternative to its larger Opus models. The company says the model can autonomously plan, use tools like browsers and terminals, and complete complex tasks that previously required more expensive systems.

Agentic capability becomes the new baseline

Sonnet 5 arrives as the AI industry shifts focus from conversational chatbots to autonomous agents. OpenAI’s GPT-5.6 Sol preview and Google’s Gemini 3.5 Flash have both emphasized agentic features, making it clear that the ability to act independently is now table stakes across all price tiers. Anthropic’s pitch for Sonnet 5 is that it delivers near-Opus-level performance at a fraction of the cost, allowing developers to balance accuracy and expense more flexibly.

Starting Tuesday, Sonnet 5 becomes the default model for free and Pro plans. Pricing is set at $2 per million input tokens and $10 per million output tokens through August 31, after which it rises to $3 and $10 respectively. This makes it cheaper than OpenAI’s GPT-5.5 and Google’s Gemini 3.1 Pro, though still more expensive than Gemini 3.5 Flash.

Performance and safety improvements

According to Anthropic, Sonnet 5 shows significant gains over its predecessor, Sonnet 4.6, in reasoning, tool use, software coding, and knowledge work. On an agentic coding benchmark, it scored 63.2%, up from 58.1% for Sonnet 4.6, while Opus 4.8 scored 69.2%. On a knowledge work benchmark, Sonnet 5 slightly outperformed Opus 4.8, which is known for handling subtle judgment calls and deep research.

Safety is also a focus. Anthropic reports lower rates of undesirable behaviors such as cooperation with misuse, deception, and sycophancy compared to Sonnet 4.6. The model is better at refusing malicious requests and resisting prompt-injection attacks. However, it does not match Opus 4.8 or the Claude Mythos Preview in handling misaligned behavior, and its ability to perform dangerous cybersecurity tasks is much lower than Opus models — a deliberate design choice for safer deployment.

Early tester feedback

Early adopters have reported that Sonnet 5 completes tasks that earlier models would abandon midway. Daniel Shepard, a senior engineer at Zapier, said the model handled a two-part job — updating Salesforce account tiers and sending launch announcements — end to end, without stalling. Fabian Hedin, co-founder of Lovable, praised the model’s ability to refuse unsafe requests cleanly and consistently, calling it as important as its ability to build.

Conclusion

Claude Sonnet 5 represents Anthropic’s bet that the next competitive frontier in AI is not just who can build the most capable agent, but who can deliver that capability at the lowest cost with the highest reliability and safety. By offering near-Opus performance at a mid-range price, Anthropic gives developers a new option to scale autonomous tasks without breaking budgets or sacrificing trust.

FAQs

Q1: How does Claude Sonnet 5 compare to OpenAI’s GPT-5.6 Sol?
Both models emphasize agentic capabilities, but Sonnet 5 is priced lower than GPT-5.6 Sol and focuses on providing near-Opus performance at a mid-range cost. OpenAI’s model allows splitting work across subagents for longer tasks, while Sonnet 5 emphasizes reliability and safety in autonomous execution.

Q2: When will Claude Sonnet 5 be available and at what price?
Sonnet 5 is available starting Tuesday as the default model for free and Pro plans. Introductory pricing is $2 per million input tokens and $10 per million output tokens through August 31, after which the input price rises to $3.

Q3: Is Claude Sonnet 5 safer than previous models?
Yes. Anthropic reports lower rates of undesirable behaviors such as deception, sycophancy, and cooperation with misuse. The model is also better at refusing malicious requests and resisting prompt-injection attacks, though it does not match Opus 4.8 in handling the most misaligned behavior.