Anthropic release Claude Opus 5, its ‘safest model yet’

Anthropic has unveiled a new AI model it says comes close to Claude Fable 5 in its levels of intelligence — at half the price.

Claude Opus 5 was just released by Anthropic on Friday. The AI company is already pushing the model out as the default model on Claude Max, Anthropic’s highest subscription plan which starts at $100 a month. It’s also the strongest available model on Claude Pro, the company’s $20-per-month subscription tier.

Claude Opus 5 tops Anthropic’s other models in key areas

Opus 5 is “designed to be used every day,” Anthropic says, adding that it’s more efficient than the company’s previous models.

Most notably, the company calls Opus 5 its “most aligned model to date.” This means Opus 5 abides by “human values” and safety guidelines.

When compared to prior models like Opus 4.8, Sonnet 5, or Fable 5, Opus 5 has the lowest rates of deceptiveness and is the least likely to be tricked into misuse, Anthropic says of its “safest model yet.”

One intriguing test highlighted by Anthropic: how fast Opus 5 solves a problem that the model has never seen before. The company says it’s three times as fast as the next-best model, and scored nearly four times higher than OpenAI’s GPT-5.6 Sol.

Claude Opus 5 benchmarks

Credit: Anthropic

Compared to Fable 5, Opus 5 performs better or the same in the vast majority of Anthropic benchmark tests. (Notable exceptions: health and legal benchmarks). It beats OpenAI’s GPT 5.6 Sol in every benchmark except for Agentic coding.

Anthropic is especially proud of Opus 5’s science benchmarks, calling the model the “most capable generally available model for scientific research.” Still, Mythos 5 beats Opus 5 in areas such as biology research.

Cybersecurity is another area where Opus 5 falls behind, and that’s by design. Anthropic says it purposefully left such tasks out of Opus 5’s training. The model can still find vulnerabilities as it expands its dataset, but is much worse at exploiting them than Mythos and previous models.

The choice is a notable one in the wake of an OpenAI agent recently going rogue and hacking a client’s server. Opus 5 was clearly trained by Anthropic with safeguards in mind.

How to try Claude Opus 5

Claude Opus 5 is available as of Friday July 24, across Anthropic’s platforms.

Opus 5 is now the default AI model on Claude Max and the strongest available model on Claude Pro. The model costs $5 per million input tokens and $25 per million output tokens, which is equivalent to Opus 4.8’s pricing.