Imagine getting almost the same performance as a top-tier AI system for half the price. That's exactly what Anthropic has achieved with its new Claude Opus 5 model. Released recently, Opus 5 offers nearly the same capabilities as the more expensive Claude Fable 5, but costs 50% less, making it an attractive choice for developers and businesses focused on coding, AI agents, and complex knowledge tasks.
Opus 5 maintains pricing at $5 per million input tokens and $25 per million output tokens, identical to its predecessor Opus 4.8 but significantly cheaper than Fable 5. Despite the lower price, Anthropic claims Opus 5 excels in performance metrics. For example, on the CursorBench 3.2 benchmark, Opus 5 scored within half a percent of Fable 5’s highest achievement while using half the budget per task. It also set new records on Frontier Bench, doubling the coding performance compared to Opus 4.8, and doing so more cost-efficiently.
Beyond coding, Opus 5 shines in knowledge work. It scored three times higher than competing models on ARC AGI 3 and outperformed rivals by 1.5 times on Zapier’s AutomationBench, all without increasing costs. Anthropic says the model excels at verifying its own work, identifying issues, and iterating to complete tasks successfully skills rare among AI models. In one scenario, Opus 5 built a computer vision pipeline independently, extracting geometrical data from images and reconstructing them into 3D FreeCAD models.
Advances reach into life sciences as well. Opus 5 outperformed Opus 4.8 by over 10 percentage points on an organic chemistry test and by nearly 8 points on a protein function task. However, it still trails behind Anthropic’s more restricted Mythos 5 model in specialized biology research and cybersecurity offensive tasks, particularly in exploit development. Opus 5 also comes equipped with safeguards to block penetration testing and exploit generation, with these protections designed to activate far less often than in past models.
Opus 5 is already live across Anthropic's platforms and the API, now the default for Claude Max users and the most capable option for Claude Pro subscribers. Developers can tap into its capabilities by calling the claude opus 5 model through the API. This launch offers a compelling package: near-best-in-class AI performance married to cost efficiency, potentially reshaping how professional AI applications are priced and accessed.
Advances in AI-driven life science tasks align with these improvements, underscoring the growing impact of such models in specialized fields.
