Anthropic has unveiled its new artificial intelligence (AI) model, Claude Opus 5, touting its cost-effectiveness.
On July 24, Anthropic introduced Opus 5 on its official website, emphasizing that it nearly matches the performance of their top public model, Fable 5, at half the cost.
Anthropic reports that Opus 5 outperformed existing public models in key benchmarks, particularly in coding and knowledge-based tasks.
In the Frontier-Bench v0.1, which evaluates coding ability in an agent terminal environment, Opus 5 scored 43.3%, surpassing OpenAI’s GPT-5.6 Sol (34.4%) and Fable 5 (33.7%).
For knowledge work assessment in GDPval-AA v2, it achieved a score of 1861, and in agent search performance, it recorded 90.8%, leading both categories.
However, Opus 5 scored 68.8% in the DeepSWE v1.1 metric for agent coding ability, slightly below GPT-5.6 Sol’s 72.7%. In the Humanity’s Last Exam (HLE) metric, which tests expert-level reasoning without external tools, it scored 56.3%, narrowly behind Preamble 5’s 56.5%.
Anthropic emphasized Opus 5’s price-to-performance ratio as its main selling point. The model’s developer usage fee remains unchanged from Opus 4.8 at 5 USD per million input tokens and 0.25 USD per output token.
Industry analysts suggest that Anthropic’s release of this cost-effective model is a strategic move to compete with high-performance, lower-priced models from rivals like OpenAI, Google, xAI, and Chinese firms Kimi K3 and DeepSeek.
Anthropic claims that Opus 5’s cybersecurity vulnerability detection capabilities are comparable to those of Mythos 5. However, they acknowledge that Opus 5 is less adept at converting detected vulnerabilities into actual cyber threats. The company also noted that Opus 5 lags behind Mythos 5 in long-term autonomous research capabilities, such as biological subfield development.