Opus 5 doubles coding performance at the same price as its predecessor

Anthropic has released Opus 5, available today across all its platforms at the same price as its predecessor: $5 per million input tokens and $25 per million output tokens.

In coding and knowledge-work evaluations such as Frontier-Bench and GDPval-AA, Opus 5 becomes the new industry benchmark. On Frontier-Bench v0.1 it scores 43.3%, against Opus 4.8's 21.1%, more than double. On CursorBench it lands within half a point of Fable 5, at half the cost per task. On novel problem-solving (ARC-AGI-3) it reaches 30.2%, far ahead of the next-best model's 7.8%. On computer-use tasks (OSWorld 2.0) it hits 70.6%, surpassing Fable 5's best result (66.1%) at less than a third of the cost per task, according to the company.

The model also improves on scientific research: it rises 10.2 percentage points over Opus 4.8 on organic chemistry, such as inferring molecular structures from spectroscopy data, and 7.7 points on protein-related tasks.

On safety, Anthropic says Opus 5 is its most aligned model to date, with the lowest rate of deceptive behavior in its internal audits. However, it notes the model does not advance the frontier in dual-use capabilities: it still trails Mythos 5 in both biological research and offensive cybersecurity. It comes close to Mythos 5 at identifying software vulnerabilities, but falls well behind at developing exploits for them.

Opus 5's safety classifiers are less restrictive than Fable 5's in cybersecurity, and Anthropic expects them to trigger around 85% less often. The model is now the default on Claude Max and the strongest available on Claude Pro, and it is also offered in Fast mode, running 2.5 times faster than standard.

Key points

• Anthropic releases Opus 5, available today at the same price as Opus 4.8.
• Doubles Opus 4.8's coding performance (Frontier-Bench).
• Nears Fable 5 on CursorBench at half the cost.
• Triples the next-best model on novel problems (ARC-AGI-3).
• Beats Fable 5 on computer use at a third of the cost.
• Improves on organic chemistry and protein-related tasks.
• Anthropic's most aligned model to date.
• Still trails Mythos 5 on biology and offensive cybersecurity.
• Safety classifiers trigger 85% less often than Fable 5's.
• Default model on Claude Max and strongest on Claude Pro.