On July 24, Anthropic released Claude Opus 5 — a new model. The main point: the price remains the same as for Opus 4.8 ($5 per million input tokens and $25 for output), but the model is noticeably stronger — on the Frontier-Bench coding benchmark it more than doubles 4.8, and on other tests it approaches the top-tier Claude Fable 5 for half its price. I am breaking down what this upgrade brings.
What is new in Opus 5 vs Opus 4.8
- Same price as Opus 4.8 ($5 / $25 per million tokens), but noticeably higher performance.
- Effort regulator (high / xhigh / max): balancing answer quality against token consumption.
- More than twice as strong as Opus 4.8 on Frontier-Bench at a lower cost per task.
- Closer to the flagship Claude Fable 5 for half its price (on CursorBench — within 0.5% of Fable 5’s peak).
- Stronger in science: +10 percentage points over 4.8 in organic chemistry.
- Default on Claude Max, the strongest model on Claude Pro.
Why this matters. “Stronger for the same money” is a direct benefit for everyone working heavily with AI: the same budget, but more gets done. And the effort regulator helps save tokens on simple tasks and max out on complex ones.
What practice will show. Benchmarks are not yet live work. Anthropic honestly states that Opus 5 is behind the Mythos 5 model in cybersecurity. And “more than doubles” on charts does not mean “twice as good” on real tasks — this will become clear in practice in a week or two.
Source: anthropic.com
