'Claude Sonnet 5' has been released, offering performance close to Opus 4.8, and is available at a low price and even on the free plan.



Anthropic released its AI model ' Claude Sonnet 5 ' on June 30, 2026 (local time). Claude Sonnet 5 is a cost-effective model that can be used even with the free plan, and in benchmark tests it has achieved scores close to those of Claude Opus 4.8.

Introducing Claude Sonnet 5 \ Anthropic

https://www.anthropic.com/news/claude-sonnet-5

The table below shows the results of various benchmark tests for 'Claude Sonnet 5,' 'Claude Sonnet 4.6,' and 'Claude Opus 4.8.' The Claude Sonnet 5 significantly outperforms the previous generation model, the Claude Sonnet 4.6, and records a score that comes close to that of the Claude Opus 4.8.



The graph below shows the test results of '

BrowseComp ,' which measures the web search capabilities of AI, with cost on the horizontal axis and score on the vertical axis. Claude Sonnet 5 can achieve high performance at a lower cost compared to Claude Sonnet 4.6. Furthermore, by increasing the cost, it can achieve a score equivalent to Claude Opus 4.8.



Similarly, '

OSWorld-Verified ,' a measure of computer usage capabilities, has shown that it enables lower-cost and higher-precision processing compared to Claude Sonnet 4.6.



Claude Sonnet 5 has a lower frequency of undesirable outputs, such as 'hallucination' and 'generation of unwanted outputs due to prompt injection,' compared to Claude Sonnet 4.6. However, the frequency is still higher compared to higher-end models like Claude Mythos Preview and Claude Opus 4.8.



Unlike Claude Mythos 5 and Claude Fable 5, Claude Sonnet 5 has not been specifically trained for cyberattack capabilities. The graph below shows the success rate of cyberattacks performed after training the AI on vulnerabilities in Firefox 147. Claude Sonnet 5 achieved a 'partial success' rate of 13.2%, where it was able to perform initial stages of attacks such as 'register manipulation,' but it never succeeded in constructing a complete attack method (exploit).



Claude Sonnet 5 is available in all plans, including the free plan, and is set as the default model in the free and Pro plans. It is also available via API. The price per 1 million tokens is $3 (approximately 488 yen) for input and $15 (approximately 2440 yen) for output, with a discounted rate of $2 (approximately 325 yen) for input and $10 (approximately 1627 yen) for output available for a limited time until August 31, 2026.

Furthermore, Claude Sonnet 5 is available for use with third-party products such as GitHub Copilot, and an API is also provided by OpenRouter.




Related Posts:

in AI, Posted by log1o_hf