⚡️ Anthropic released Claude Opus 5 - almost the level of their top model at half the price.
Opus 5 costs the same as Opus 4.8 ($5 per million input tokens, $25 for output), but in terms of intelligence, it nearly reaches Fable 5, which is twice as expensive. On CursorBench 3.2 with maximum effort, the difference with Fable 5 is 0.5 percentage points, at half the price. On Frontier-Bench v0.1, Opus 5's result is twice that of Opus 4.8.
On ARC-AGI 3, the model scored three times more points than the next strongest model. On OSWorld 2.0, it surpasses all models in its price range and exceeds Fable 5, which costs three times more. In scientific tasks, it shows a plus of 10.2 percentage points in organic chemistry and 7.7 in protein analysis compared to Opus 4.8.
Anthropic deliberately does not position Opus 5 as state-of-the-art in dual-use tasks - cybersecurity and biology. There, Mythos 5 will remain with stricter access restrictions - the company clearly distinguishes between a "model for work" and a "model that is scary to give to everyone."
The model has already become the base in Claude Max and the most powerful available in Claude Pro, operating in API, claude.ai, Claude Code, and Claude Cowork. Cursor describes it as "Fable 5 level intelligence at the speed and price of Opus." Zapier achieved 100% workflow completion in customer retention, which previously failed completely.
https://x.com/claudeai/status/2080699495453528290
#anthropic@rvnikita_blog #claude@rvnikita_blog #ai@rvnikita_blog #llm@rvnikita_blog