Anthropic launches Claude Opus 5 with better performance at same price

Anthropic has released Claude Opus 5, the latest flagship model in the Opus series, offering better performance at the same price. It's designed for everyday high-performance use, with capabilities close to the higher-end Claude Fable 5 but at about half the cost. It's now the default model for Claude Max.

Opus 5 leads in code and knowledge work benchmarks like Frontier-Bench, doubling task performance over Opus 4.8 while lowering per-task costs. It also shows strong gains in scientific research, such as inferring molecular structures from spectral data, scoring 10.2 percentage points higher.

In real-world tests, Opus 5 excels at self-checking and iterating. It created a 3D model from machine part drawings, found and fixed a bug in an open-source package manager, and helped an engineer build a trading data interface. It has fewer safety restrictions than Fable 5, with security triggers expected to drop by about 85%.

Pricing remains the same: $5 per million input tokens and $25 per million output tokens. Anthropic also introduced Opus 5 Fast mode, running about 2.5 times faster at double the cost, along with new test features for tool switching and automatic fallback in the API.