Anthropic Launches Claude Opus 5, Offering Top-Tier AI Intelligence at Half Price

Artificial intelligence company Anthropic has launched Claude Opus 5, a new model designed to bridge the gap between high-end performance and affordability. The company claims Opus 5 achieves intelligence levels close to its most advanced model, Claude Fable 5, while costing significantly less.

Announced via Anthropic's official blog, Opus 5 is described as a "thoughtful and proactive" model. It has set new performance records on benchmarks like Frontier-Bench and GDPval-AA for coding and knowledge work. However, Anthropic notes it does not match the cybersecurity capabilities of its top-tier model, Claude Mythos 5.

The pricing strategy remains the same as the previous Opus 4.8 model, with costs set at $5 per million input tokens and $25 per million output tokens. This means users receive enhanced performance at the same price point, positioning Opus 5 as the most powerful model available through Claude Max and Claude Pro subscriptions.

“

Claude Opus 5 provides Fable 5-level intelligence at Opus-level speed and cost. It's right below Fable 5 on CursorBench and behaves very similarly.

Performance improvements are evident across several benchmarks. Opus 5 reportedly surpassed all other models on Frontier-Bench v0.1, delivering double the performance of Opus 4.8 at a lower cost per task. It closely approached Fable 5's score on CursorBench 3.2 with significantly reduced costs. In the ARC-AGI 3 test for new problem-solving abilities, it scored three times higher than the next-best model. It also showed a 1.5 times higher success rate on Zapier AutomationBench tasks compared to the previous model at the same cost.

Anthropic highlighted real-world applications, including an instance where Opus 5 autonomously created a computer vision pipeline to reconstruct a 3D model from pixel data alone, a task competitors failed. In another case, it identified and fixed a deep-seated bug in an open-source package that community patches had missed. Partners like Cursor and Zapier have praised Opus 5 for its speed, intelligence, and cost-effectiveness, noting its ability to handle complex tasks efficiently.

Beyond performance, Anthropic emphasized Opus 5's safety features. Internal testing indicated it has the highest alignment among Anthropic models based on automated behavior audits, with low rates of deceptive behavior and misuse vulnerability. While capable of detecting code vulnerabilities, its design prevents it from exploiting them, balancing defensive research support with the prevention of offensive tool creation.

“

Opus 5 reached number 1 on Zapier's AutomationBench leaderboard without using more tokens than previous Claude models. It handled the entire account-churn prevention task, which previous models failed.