Claude Sonnet 5 Launches, Anthropic Warms Up AI Agent Price War

Anthropic released Claude Sonnet 5, a mid-range AI model designed to run more independent tasks at a lower cost. This launch underscores a new direction in AI competition. It's no longer just about who is the smartest, but who is the most efficient and stable in running AI agents.

Launching a report from TechCrunch, Wednesday, July 1, Claude Sonnet 5 began to become the default model for free and Pro package users and is available to all customers starting Tuesday.

This model is positioned as a more powerful version of Sonnet 4.6. Anthropic says Sonnet 5 is capable of making plans, using tools such as a browser and terminal, and working autonomously at a level that a few months ago still required a larger and more expensive model.

In the world of AI, such capabilities are commonly referred to as agentic AI, that is, AI that not only answers questions, but can also carry out a series of tasks more independently.

Price is Anthropic's main weapon. As of August 31, Sonnet 5 was priced at US$2 per one million input tokens and US$10 per one million output tokens. After that, the rate rose to US$3 per one million input tokens and US$15 per one million output tokens.

Tokens are units of text that are used to calculate the cost of using an AI model. The more text that is processed or generated, the higher the cost.

That price makes the Sonnet 5 cheaper than OpenAI's Opus 4.8, GPT-5.5, and Google's Gemini 3.1 Pro. However, this model is still more expensive than the Gemini 3.5 Flash.

Anthropic claims that Sonnet 5's performance is close to Opus 4.8, but at a lower cost. On the agenttic programming benchmark, Sonnet 5 scored 63.2 percent. That figure is below Opus 4.8's 69.2 percent, but above Sonnet 4.6's 58.1 percent.

On the knowledge-based work benchmark, Sonnet 5 even slightly outperforms Opus 4.8, a model known for its strength in deep research and complex decision-making.

"Opus 4.8 is still the choice for higher accuracy, but Sonnet 5 gives developers a cheaper option with much better quality than before," wrote Anthropic.

According to TechCrunch, Anthropic's move is in line with the direction of OpenAI and Google. OpenAI has just launched GPT-5.6 Sol in a preview version, while Google previously introduced Gemini 3.5 Flash as a model more suitable for agenttic tasks.

In terms of security, Sonnet 5 is claimed to be better than its predecessor. This model is said to be more capable of rejecting malicious requests, more resistant to prompt injection attacks, and less likely to hallucinate or follow the user's will too much.

Prompt injection is an attempt to sneak in hidden commands so that AI ignores safety rules.

Even so, Sonnet 5 is not yet at the level of Opus 4.8 and Claude Mythos Preview in terms of inconsistent behavior. Anthropic also said that Sonnet 5's ability to perform dangerous cyber security tasks is much lower than the current Opus model.

The launch of Sonnet 5 shows that the competition for AI models is now increasingly based on cost and reliability. For developers, a model that is cheap but powerful enough can be an important choice for running daily automation without burdening the budget.