The 'Claude Opus 5' has been released, offering comparable performance to the Claude Fable 5 at half the price, with fewer restrictions.



Anthropic released the ' Claude Opus 5 ' in the early hours of July 25, 2026, Japan time. It is touted as offering performance equivalent to the Claude Fable 5 at half the price, and third-party tests have confirmed that it outperforms the Claude Fable 5.

Introducing Claude Opus 5 \ Anthropic

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




Anthropic offers four consumer-向け AI models: the fast 'Haiku,' the balanced 'Sonnet,' the high-performance 'Opus,' and the top-performing 'Fable.' The newly released 'Claude Opus 5' is being promoted as offering performance equivalent to the Claude Fable 5 at the same price as the Claude Opus 4.8.

The table below shows the benchmark results for 'Claude Opus 5,' 'Claude Fable 5,' 'Claude Opus 4.8,' and 'GPT-5.6 Sol.' The Claude Opus 5 outperforms the Claude Fable 5 in several tests.



The following are the results of Frontier-Bench v0.1, which measures performance as a coding agent AI. Claude Opus 5 achieves a higher score at a lower price than Claude Fable 5.



The independent organization

Artificial Analysis has also published its performance measurement results for the Claude Opus 5. In terms of intelligence performance, it was rated as the world's best model, surpassing the Claude Fable 5 by one point.



Ranked #1 in coding performance as well.



It also ranks first in agent performance.



The graph below shows the cost per task on the horizontal axis and intelligence performance on the vertical axis. Claude Opus 5 was rated as being cheaper and more powerful than Claude Fable 5.



The results of

ARC-AGI-3 , which measures the intelligence level of AI, are shown below. It shows a significant difference compared to Gemini 3.1 Pro and GPT-5.6.



Claude Opus 5 is also characterized by its high level of safety. The graph below shows the rate of unexpected AI behavior. Claude Opus 5 has a lower rate of unexpected behavior than Claude Opus 4.8 and Claude Mythos 5.



Claude Opus 5 intentionally avoids learning tasks related to cyberattacks, and its cybersecurity capabilities are a step below those of Claude Fable 5. The results of Anthropic's proprietary test, 'OSS-Fuzz,' are as follows: While it scored close to Claude Mythos 5 in tests that discover vulnerabilities in open-source projects, it significantly underperformed Claude Mythos 5 in terms of the success rate of attacks exploiting those vulnerabilities.



Claude Opus 5 includes safeguards against cyberattacks and biological weapons; if these safeguards are triggered, the process will fall back to Claude Opus 4.8. The safeguards are less stringent than those in Claude Fable 5, resulting in approximately 85% fewer interventions compared to Claude Fable 5.

Claude Opus 5 is available to subscribers of paid subscription plans and is the default model in the Max plan. The API fees are the same as Claude Opus 4.8, with a fee of $5 (approximately 819 yen) per million tokens for input and $25 (approximately 4096 yen) for output.

Related Posts:

in AI, Posted by log1o_hf