Anthropic just released Claude Opus 5, and the pitch is simple: it comes close to the frontier intelligence of Claude Fable 5 at half the price. It is available today, and it is already the new state of the art on coding and knowledge work.
More than double the old Opus
On Frontier-Bench v0.1, the agentic terminal coding benchmark, Opus 5 scores 43.3%. That is more than double Opus 4.8 at 21.1%, and it clears both GPT-5.6 Sol at 34.4% and Fable 5 at 33.7%. One generation, and the coding number roughly doubled.
It triples the field on brand new puzzles
The most striking chart is ARC-AGI-3, a set of novel reasoning puzzles the models have never seen before. Opus 5 on high effort scores around 30, roughly triple the next best model and far beyond where Opus 4.8 lands. This is the test that rewards genuine reasoning over memorization, and Opus 5 pulls away from the entire field.
Frontier brains, half the cost
Here is where the title comes from. Across coding (Frontier-Bench), computer use (OSWorld 2.0), and Cursor's own CursorBench, Opus 5's line sits at the top while costing less per task than the models it beats. On OSWorld 2.0 it reaches about 70%, ahead of Fable 5, GPT-5.6 Sol, and Opus 4.8. You get frontier level intelligence without paying frontier level prices, which is exactly what makes it usable every day.
The one place it is not first
To be fair, Opus 5 is not the best at everything. Anthropic says it still trails Claude Mythos 5 on cybersecurity tasks. Everywhere else in the coding and knowledge work lineup, though, it is the new state of the art.
How to use Claude Opus 5 today
Opus 5 is available right now. It is the new default model on Claude Max and the strongest model on Claude Pro, and it is built to be your everyday driver, not just a benchmark hero. There is also a new fast mode for when you want the answer quicker. If you live in Claude Code or run agents all day, this is the upgrade to switch to.